diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/benchmark_start.sh b/metro-ai-suite/metro-vision-ai-app-recipe/benchmark_start.sh
new file mode 100755
index 000000000..9031cfd38
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/benchmark_start.sh
@@ -0,0 +1,326 @@
+#!/bin/bash
+
+DLSPS_NODE_IP="localhost"
+
+function get_pipeline_status() {
+    curl -k -s "https://$DLSPS_NODE_IP/api/pipelines/status"
+}
+
+function run_pipelines() {
+  local num_pipelines=$1
+  local payload_data=$2
+  local pipeline_name=$3
+
+  echo
+  echo -n ">>>>> Initialization: Starting $num_pipelines pipeline(s) of type '$pipeline_name'..."
+  
+  for (( x=1; x<=num_pipelines; x++ )); do
+    response=$(curl -k -s "https://$DLSPS_NODE_IP/api/pipelines/user_defined_pipelines/${pipeline_name}" \
+      -X POST -H "Content-Type: application/json" -d "$payload_data")
+    
+    if [ $? -ne 0 ] || [[ "$response" == *"Error"* ]]; then
+      echo -e "\nError: curl command failed or pipeline returned error. Check the deployment status."
+      echo "Response: $response"
+      return 1
+    fi
+    sleep 1 # Brief pause between requests
+  done
+  
+  # Wait for all pipelines to be in RUNNING state
+  echo -n ">>>>> Waiting for pipelines to initialize..."
+  local running_count=0
+  local attempts=0
+  while [ "$running_count" -lt "$num_pipelines" ] && [ "$attempts" -lt 60 ]; do
+    status_output=$(get_pipeline_status)
+    running_count=$(echo "$status_output" | jq '[.[] | select(.state=="RUNNING")] | length')
+    
+    echo -n "."
+    attempts=$((attempts + 1))
+    sleep 2
+  done
+  
+  if [ "$running_count" -ge "$num_pipelines" ]; then
+    echo " All pipelines are running."
+    return 0
+  else
+    echo " Error: Not all pipelines entered RUNNING state."
+    get_pipeline_status | jq
+    return 1
+  fi
+}
+
+function stop_all_pipelines() {
+  echo
+  echo ">>>>> Attempting to stop all running pipelines."
+  
+  local pipelines_str
+  pipelines_str=$(get_pipeline_status | jq -r '[.[] | select(.state=="RUNNING") | .id] | join(",")')
+  
+  if [ $? -ne 0 ]; then
+    echo -e "\nError: Failed to get pipeline status."
+    return 1
+  fi
+
+  if [ -z "$pipelines_str" ]; then
+    echo "No running pipelines found."
+    return 0
+  fi
+
+  IFS=',' read -ra pipelines <<< "$pipelines_str"
+  
+  echo "Found ${#pipelines[@]} running pipelines to stop."
+
+  for pipeline_id in "${pipelines[@]}"; do
+    curl -k -s --location -X DELETE "https://$DLSPS_NODE_IP/api/pipelines/${pipeline_id}" &
+  done
+  
+  wait
+  echo "All stop requests sent."
+  unset IFS
+
+  echo -n ">>>>> Waiting for all pipelines to stop..."
+  local running=true
+  while $running; do
+    echo -n "."
+    local status
+    status=$(get_pipeline_status | jq '.[] | .state' | grep "RUNNING")
+    if [[ -z "$status" ]]; then
+      running=false
+    else
+      sleep 3
+    fi
+  done
+  echo " done."
+  echo
+  return 0
+}
+
+declare -gA final_avg_fps
+
+function check_all_streams_meet_target_fps() {
+    local duration=$1
+    local target_fps=$2
+
+    # Get initial list of running pipeline IDs
+    local pipeline_ids
+    pipeline_ids=$(get_pipeline_status | jq -r '[.[] | select(.state=="RUNNING") | .id] | .[]')
+
+    if [ -z "$pipeline_ids" ]; then
+        echo "No running streams to monitor."
+        return 1 # Fail if no streams are running
+    fi
+
+    declare -A fps_sums
+    declare -A sample_counts
+    unset final_avg_fps
+    declare -gA final_avg_fps
+
+    # Initialize sums and counts for each pipeline
+    for id in $pipeline_ids; do
+        fps_sums[$id]=0
+        sample_counts[$id]=0
+    done
+
+    echo ">>>>> Monitoring FPS for $duration seconds..."
+    local start_time=$SECONDS
+    while (( SECONDS - start_time < duration )); do
+        local elapsed_time=$((SECONDS - start_time))
+        echo -ne "Monitoring... ${elapsed_time}s / ${duration}s\r"
+
+        local status_output
+        status_output=$(get_pipeline_status)
+
+        for id in $pipeline_ids; do
+            # Extract avg_fps for the specific pipeline ID
+            local current_fps
+            current_fps=$(echo "$status_output" | jq -r --arg ID "$id" '.[] | select(.id==$ID) | .avg_fps')
+
+            if [ -n "$current_fps" ] && [[ "$current_fps" != "null" ]]; then
+                fps_sums[$id]=$(echo "${fps_sums[$id]} + $current_fps" | bc)
+                sample_counts[$id]=$((sample_counts[$id] + 1))
+            fi
+        done
+        sleep 2
+    done
+    echo -ne "\n" # Move to next line after progress bar finishes
+
+    # Now, check if the average of each stream met the target
+    local all_streams_met_target=true
+    for id in $pipeline_ids; do
+        local num_samples=${sample_counts[$id]}
+        if [ "$num_samples" -gt 0 ]; then
+            local total_fps=${fps_sums[$id]}
+            local avg_fps
+            avg_fps=$(echo "scale=2; $total_fps / $num_samples" | bc)
+            final_avg_fps[$id]=$avg_fps
+            
+            echo "Stream $id Average FPS: $avg_fps"
+
+            if (( $(echo "$avg_fps < $target_fps" | bc -l) )); then
+                echo "  -> ❌ FAILED to meet target FPS of $target_fps"
+                all_streams_met_target=false
+            else
+                echo "  -> ✅ OK"
+            fi
+        else
+            echo "Stream $id: No FPS data collected."
+            all_streams_met_target=false
+        fi
+    done
+
+    if $all_streams_met_target; then
+        return 0 # Success
+    else
+        return 1 # Failure
+    fi
+}
+
+function run_stream_density_mode() {
+    local payload_file=$1
+    local target_fps=$2
+    local duration=$3
+    
+    echo ">>>>> Running in Stream-Density Calculation Mode (Target FPS: $target_fps)"
+    
+    local optimal_streams=0
+    local current_streams=1
+    declare -A last_successful_fps
+    
+    # Extract pipeline name and payload body from the JSON file
+    local pipeline_name
+    pipeline_name=$(jq -r '.[0].pipeline' "$payload_file")
+    local payload_body
+    payload_body=$(jq '.[0].payload' "$payload_file")
+
+    if [ -z "$pipeline_name" ] || [ -z "$payload_body" ]; then
+        echo "Error: Could not extract 'pipeline' or 'payload' from $payload_file"
+        exit 1
+    fi
+
+    while true; do
+        echo
+        echo "--- Testing with $current_streams stream(s) ---"
+        
+        run_pipelines "$current_streams" "$payload_body" "$pipeline_name"
+        if [ $? -ne 0 ]; then
+            echo "Failed to start pipelines. Aborting."
+            break
+        fi
+        
+        echo ">>>>> Waiting 10 seconds for stabilization..."
+        sleep 10
+        
+        if check_all_streams_meet_target_fps "$duration" "$target_fps"; then
+            echo "✓ Target FPS met with $current_streams stream(s)."
+            optimal_streams=$current_streams
+            
+            # Save the FPS values from this successful run
+            unset last_successful_fps
+            declare -A last_successful_fps
+            for id in "${!final_avg_fps[@]}"; do
+                last_successful_fps[$id]=${final_avg_fps[$id]}
+            done
+            
+            stop_all_pipelines
+            sleep 5
+            
+            current_streams=$((current_streams + 1))
+        else
+            echo "❌ Target FPS not met with $current_streams stream(s)."
+            break
+        fi
+    done
+    
+    stop_all_pipelines
+    
+    echo
+    echo "======================================================"
+    if [ "$optimal_streams" -gt 0 ]; then
+        echo "✅ FINAL RESULT: Stream-Density Benchmark Completed!"
+        echo "   Maximum $optimal_streams stream(s) can achieve the target FPS of $target_fps."
+        echo
+        echo "   Average FPS per stream for the optimal configuration:"
+        for id in "${!last_successful_fps[@]}"; do
+            echo "     - Stream $id: ${last_successful_fps[$id]} FPS"
+        done
+    else
+        echo "❌ FINAL RESULT: Target FPS Not Achievable."
+        echo "   No configuration could achieve the target FPS of $target_fps."
+    fi
+    echo "======================================================"
+}
+
+# --- Main Script ---
+
+function usage() {
+    echo "Usage: $0 -p <payload_file> [-n <num_pipelines> | -t <target_fps>] [-i <interval>]"
+    echo
+    echo "Modes:"
+    echo "  Fixed Stream Mode: Provide -n to run a specific number of pipelines."
+    echo "  Stream-Density Mode: Omit -n and provide -t to find the optimal number of streams."
+    echo
+    echo "Arguments:"
+    echo "  -p <payload_file>    : (Required) Path to the benchmark payload JSON file."
+    echo "  -n <num_pipelines>   : Number of pipelines to run."
+    echo "  -t <target_fps>      : Target FPS for stream-density mode (default: 28.5)."
+    echo "  -i <interval>        : Monitoring interval in seconds for stream-density mode (default: 60)."
+    exit 1
+}
+
+num_pipelines=""
+payload_file=""
+target_fps="28.5"
+interval=60
+
+while getopts "n:p:t:i:" opt; do
+  case ${opt} in
+    n ) num_pipelines=$OPTARG ;;
+    p ) payload_file=$OPTARG ;;
+    t ) target_fps=$OPTARG ;;
+    i ) interval=$OPTARG ;;
+    \? ) usage ;;
+  esac
+done
+
+if [ -z "$payload_file" ]; then
+    echo "Error: Payload file is required."
+    usage
+fi
+
+if [ ! -f "$payload_file" ]; then
+    echo "Error: Benchmark payload file not found: $payload_file"
+    exit 1
+fi
+
+stop_all_pipelines
+if [ $? -ne 0 ]; then
+   exit 1
+fi
+
+if [ -n "$num_pipelines" ]; then
+    if ! [[ "$num_pipelines" =~ ^[0-9]+$ ]] || [ "$num_pipelines" -le 0 ]; then
+        echo "Error: Number of pipelines (-n) must be a positive integer."
+        usage
+    fi
+    
+    pipeline_name=$(jq -r '.[0].pipeline' "$payload_file")
+    payload_body=$(jq '.[0].payload' "$payload_file")
+
+    if [ -z "$pipeline_name" ] || [ -z "$payload_body" ]; then
+        echo "Error: Could not extract 'pipeline' or 'payload' from $payload_file"
+        exit 1
+    fi
+    
+    run_pipelines "$num_pipelines" "$payload_body" "$pipeline_name"
+    if [ $? -ne 0 ]; then
+      exit 1
+    fi
+    
+    echo
+    echo ">>>>> $num_pipelines pipeline(s) are running."
+    echo ">>>>> Results can be visualized in Grafana at 'https://localhost/grafana'"
+    echo ">>>>> Pipeline status can be checked with 'curl -k https://localhost/api/pipelines/status'"
+
+else
+    run_stream_density_mode "$payload_file" "$target_fps" "$interval"
+fi
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/README.md b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/README.md
index 1629ed847..715b41ad7 100644
--- a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/README.md
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/README.md
@@ -36,4 +36,5 @@ It integrates various components such as MQTT, Node-RED, and Grafana to provide
 - [System Requirements](./docs/user-guide/system-requirements.md): Check the hardware and software requirements for deploying the application.
 - [Get Started](./docs/user-guide/get-started.md): Follow step-by-step instructions to set up the application.
 - [How to Deploy with Helm](./docs/user-guide/how-to-deploy-with-helm.md): How to deploy the application using Helm on a Kubernetes cluster.
+- [How to Run Benchmarking](./docs/user-guide/how-to-benchmark.md): A guide on how to benchmark the application.
 - [Support and Troubleshooting](./docs/user-guide/support.md): Find solutions to common issues and troubleshooting steps.
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_cpu_payload.json b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_cpu_payload.json
new file mode 100644
index 000000000..0a3f08089
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_cpu_payload.json
@@ -0,0 +1,18 @@
+[
+    {
+        "pipeline": "object_tracking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/intel/pedestrian-and-vehicle-detector-adas-0001/FP16-INT8/pedestrian-and-vehicle-detector-adas-0001.xml",
+                    "device": "CPU",
+                    "inference-interval": 3,
+                    "inference-region": 0,
+                    "pre-process-backend": "opencv",
+                    "threshold": 0.7,
+                    "model-instance-id": "instcpu0"
+                }
+            }
+        }
+    }
+]
\ No newline at end of file
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_gpu_payload.json b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_gpu_payload.json
new file mode 100644
index 000000000..731e3a949
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_gpu_payload.json
@@ -0,0 +1,21 @@
+[
+    {
+        "pipeline": "object_tracking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/intel/pedestrian-and-vehicle-detector-adas-0001/FP16-INT8/pedestrian-and-vehicle-detector-adas-0001.xml",
+                    "device": "GPU",
+                    "inference-interval": 5,
+                    "inference-region": 0,
+                    "batch-size": 8,
+                    "nireq": 2,
+                    "ie-config": "NUM_STREAMS=2",
+                    "pre-process-backend": "va-surface-sharing",
+                    "threshold": 0.7,
+                    "model-instance-id": "instgpu0"
+                }
+            }
+        }
+    }
+]
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_npu_payload.json b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_npu_payload.json
new file mode 100644
index 000000000..803e478a6
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/benchmark_npu_payload.json
@@ -0,0 +1,20 @@
+[
+    {
+        "pipeline": "object_tracking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/intel/pedestrian-and-vehicle-detector-adas-0001/FP16-INT8/pedestrian-and-vehicle-detector-adas-0001.xml",
+                    "device": "NPU",
+                    "inference-interval": 5,
+                    "inference-region": 0,
+                    "batch-size": 1,
+                    "nireq": 4,
+                    "pre-process-backend": "va-surface-sharing",
+                    "threshold": 0.7,
+                    "model-instance-id": "instnpu0"
+                }
+            }
+        }
+    }
+]
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/how-to-benchmark.md b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/how-to-benchmark.md
new file mode 100644
index 000000000..7c35db83c
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/how-to-benchmark.md
@@ -0,0 +1,135 @@
+# How to Run Benchmarking
+
+This guide provides step-by-step instructions for running the benchmarking script to evaluate the performance of the Loitering Detection application. The script can help you determine the maximum number of concurrent video streams your system can handle while meeting a specific performance target (e.g., frames per second).
+
+## Overview of the Benchmarking Script
+
+The `benchmark_start.sh` script, located in the `metro-vision-ai-app-recipe` directory, automates the process of running performance tests on the DL Streamer Pipeline Server. It offers two primary modes of operation:
+
+*   **Fixed Stream Mode:** Runs a specified number of concurrent video processing pipelines. This mode is useful for testing a specific, known workload.
+*   **Stream-Density Mode:** Automatically determines the maximum number of streams that can be processed while maintaining a target Frames Per Second (FPS). This is ideal for capacity planning and finding the performance limits of your hardware.
+
+## Prerequisites
+
+Before running the benchmarking script, ensure you have the following:
+
+*   A successful deployment of the Loitering Detection application using Helm, as described in the [How to Deploy with Helm](./how-to-deploy-with-helm.md) guide.
+*   The `jq` command-line JSON processor must be installed. You can install it on Ubuntu with:
+    ```bash
+    sudo apt-get update && sudo apt-get install -y jq
+    ```
+*   The `benchmark_start.sh` script and a payload configuration file must be available on your system.
+
+## Understanding the Payload File
+
+The benchmarking script requires a JSON payload file to configure the pipelines that will be tested. These payload files are located within each sample application's directory (e.g., `loitering-detection/`). The script uses this file to specify the pipeline to run and the configuration for the video source, destination, and parameters.
+
+Here is an example of a payload file, `benchmark_cpu_payload.json`:
+
+```json
+[
+    {
+        "pipeline": "object_tracking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/intel/pedestrian-and-vehicle-detector-adas-0001/FP16-INT8/pedestrian-and-vehicle-detector-adas-0001.xml",
+                    "device": "CPU",
+                    "inference-interval": 3,
+                    "threshold": 0.7
+                }
+            }
+        }
+    }
+]
+```
+
+*   `pipeline`: The name of the pipeline to execute (e.g., `object_tracking_benchmarking`).
+*   `payload`: An object containing the configuration for the pipeline instance.
+    *   `parameters`: Allows you to set pipeline-specific parameters, such as the `detection-device` (CPU, GPU, or NPU) and other model-related properties.
+
+## Step 1: Configure the Benchmarking Script
+
+Before running the script, you may need to adjust the `DLSPS_NODE_IP` variable within `benchmark_start.sh` if your DL Streamer Pipeline Server is not running on `localhost`.
+
+```bash
+# Edit the benchmark_start.sh script if needed
+nano benchmark_start.sh
+```
+
+Change `DLSPS_NODE_IP="localhost"` to the correct IP address of the node where the service is exposed.
+
+## Step 2: Run the Benchmarking Script
+
+The script can be run in two different modes.
+
+Navigate to the `metro-vision-ai-app-recipe` directory to run the script.
+
+```bash
+cd edge-ai-suites/metro-ai-suite/metro-vision-ai-app-recipe/
+```
+
+### Fixed Stream Mode
+
+In this mode, you specify the exact number of pipelines to run concurrently. This is useful for simulating a known workload.
+
+**To run 4 pipelines simultaneously using the CPU payload:**
+
+```bash
+./benchmark_start.sh -p loitering-detection/benchmark_cpu_payload.json -n 4
+```
+
+*   `-p loitering-detection/benchmark_cpu_payload.json`: Specifies the path to your payload configuration file.
+*   `-n 4`: Sets the number of concurrent pipelines to run.
+
+The script will start the 4 pipelines and print their status. You can then monitor their performance via Grafana, by using the pipeline status API endpoint, or by running the `./sample_status.sh` script from the `metro-vision-ai-app-recipe` directory to check the average FPS.
+
+### Stream-Density Mode
+
+In this mode, the script automatically finds the maximum number of streams that can run while maintaining a target FPS. This is useful for determining the capacity of your system.
+
+**To find the maximum number of streams that can achieve at least 28.5 FPS on CPU:**
+
+```bash
+./benchmark_start.sh -p loitering-detection/benchmark_cpu_payload.json -t 28.5
+```
+
+*   `-p loitering-detection/benchmark_cpu_payload.json`: Specifies the path to your payload configuration file.
+*   `-t 28.5`: Sets the target average FPS per stream. The default is `28.5`.
+*   `-i 60`: (Optional) Sets the monitoring interval in seconds for collecting FPS data. The default is `60`.
+
+The script will start with one stream, measure the FPS, and if the target is met, it will stop, add another stream, and repeat the process. This continues until the average FPS drops below the target. The script will then report the maximum number of streams that successfully met the performance goal.
+
+**Example Output:**
+
+```
+======================================================
+✅ FINAL RESULT: Stream-Density Benchmark Completed!
+   Maximum 4 stream(s) can achieve the target FPS of 28.5.
+======================================================
+```
+
+### How Stream Performance is Evaluated
+
+In Stream-Density Mode, the script evaluates if the system can sustain a target FPS across all concurrent streams. The process is as follows:
+
+1.  **Individual Stream Monitoring:** The script monitors each running pipeline instance (stream) independently.
+2.  **Sampling:** For the duration of the monitoring interval (e.g., 60 seconds), it samples the `avg_fps` value from each stream every 2 seconds.
+3.  **Averaging per Stream:** After the interval, it calculates the average FPS for *each stream* based on the samples collected for that specific stream.
+4.  **Validation:** The performance goal is considered met only if **every single stream's** calculated average FPS is greater than or equal to the target FPS. If even one stream falls below the target, the test fails for that number of concurrent streams.
+
+This ensures that the reported optimal stream count represents a stable configuration where all streams are performing adequately, rather than relying on a combined average that could hide underperforming streams.
+
+## Step 3: Stop the Benchmarking
+
+To stop the pipelines, run the `sample_stop.sh` script from the `metro-vision-ai-app-recipe` directory.
+
+```bash
+./sample_stop.sh
+```
+
+This script will stop all running pipelines that were initiated by the benchmark script.
+
+## Summary
+
+In this guide, you learned how to use the `benchmark_start.sh` script to run performance tests on your Loitering Detection application. You can now measure performance for a fixed number of streams or automatically determine the maximum stream density your system can support.
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/index.rst b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/index.rst
index 9f3ed2cb0..2587a6e4c 100644
--- a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/index.rst
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/docs/user-guide/index.rst
@@ -19,8 +19,10 @@ By utilizing cutting-edge technologies and pre-trained deep learning models, thi
 
    how-to-customize-application
    how-to-deploy-with-helm
+   how-to-benchmark
    how-to-deploy-with-edge-orchestrator
 
+
 .. toctree::
    :caption: Source Code
 
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/src/dlstreamer-pipeline-server/config.json b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/src/dlstreamer-pipeline-server/config.json
index 27d162392..c0493b397 100644
--- a/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/src/dlstreamer-pipeline-server/config.json
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/loitering-detection/src/dlstreamer-pipeline-server/config.json
@@ -119,23 +119,25 @@
                 },
                 "auto_start": false,
                 "publish_frame": true
-            }
-        ]
-    },
-    "interfaces": {
-        "Publishers": [
+            },
             {
-                "Name": "default",
-                "Type": "zmq_ipc",
-                "zmq_send_hwm": 50,
-                "EndPoint": "/EII/sockets",
-                "Topics": [
-                    "yolov5",
-                    "yolov5_effnet"
-                ],
-                "AllowedClients": [
-                    "*"
-                ]
+                "name": "object_tracking_benchmarking",
+                "source": "gstreamer",
+                "queue_maxsize": 50,
+                "pipeline": "multifilesrc location=/home/pipeline-server/videos/VIRAT_S_000101.mp4 loop=true ! parsebin ! vah264dec ! vapostproc ! video/x-raw(memory:VAMemory) ! gvaattachroi roi=0,200,300,400 ! gvadetect name=detection ! queue ! gvafpscounter ! appsink sync=false async=false",
+                "description": "Object detection benchmarking pipeline",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "detection-properties": {
+                            "element": {
+                                "name": "detection",
+                                "format": "element-properties"
+                            }
+                        }
+                    }
+                },
+                "auto_start": false
             }
         ]
     }
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/README.md b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/README.md
index f9eb217ca..46b54a8b6 100644
--- a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/README.md
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/README.md
@@ -54,4 +54,5 @@ for display in Grafana.
 
 - [Release Notes](./docs/user-guide/release-notes.md)
 - [How to customize the application](./docs/user-guide/how-to-customize-application.md)
+- [How to Run Benchmarking](./docs/user-guide/how-to-benchmark.md): A guide on how to benchmark the application.
 - [Support and Troubleshooting](./docs/user-guide/support.md)
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_cpu_payload.json b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_cpu_payload.json
new file mode 100644
index 000000000..d8fdab913
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_cpu_payload.json
@@ -0,0 +1,25 @@
+[
+    {
+        "pipeline": "smart_parking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/public/yolov10s/FP16/yolov10s.xml",
+                    "device": "CPU",
+                    "model-instance-id": "inst0",
+                    "inference-interval": 3,
+                    "inference-region": 0,
+                    "pre-process-backend": "opencv",
+                    "threshold": 0.7
+                },
+                "classification-properties": {
+                    "model": "/home/pipeline-server/models/colorcls2/colorcls2.xml",
+                    "device": "CPU",
+                    "model-instance-id": "inst1",
+                    "inference-interval": 3,
+                    "pre-process-backend": "opencv"
+                }
+            }
+        }
+    }
+]
\ No newline at end of file
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_gpu_npu_payload.json b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_gpu_npu_payload.json
new file mode 100644
index 000000000..7c836971c
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_gpu_npu_payload.json
@@ -0,0 +1,30 @@
+[
+    {
+        "pipeline": "smart_parking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/public/yolov10s/FP16/yolov10s.xml",
+                    "device": "GPU",
+                    "batch-size": 8,
+                    "model-instance-id": "instgpu0",
+                    "inference-interval": 5,
+                    "inference-region": 0,
+                    "nireq": 2,
+                    "ie-config": "NUM_STREAMS=2",
+                    "pre-process-backend": "va-surface-sharing",
+                    "threshold": 0.7
+                },
+                "classification-properties": {
+                    "model": "/home/pipeline-server/models/colorcls2/colorcls2.xml",
+                    "device": "NPU",
+                    "model-instance-id": "instnpu0",
+                    "inference-interval": 1,
+                    "batch-size": 1,
+                    "nireq": 2,
+                    "pre-process-backend": "va-surface-sharing"
+                }
+            }
+        }
+    }
+]
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_gpu_payload.json b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_gpu_payload.json
new file mode 100644
index 000000000..2c493c220
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/benchmark_gpu_payload.json
@@ -0,0 +1,31 @@
+[
+    {
+        "pipeline": "smart_parking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/public/yolov10s/FP16/yolov10s.xml",
+                    "device": "GPU",
+                    "batch-size": 8,
+                    "model-instance-id": "instgpu0",
+                    "inference-interval": 5,
+                    "inference-region": 0,
+                    "nireq": 2,
+                    "ie-config": "NUM_STREAMS=2",
+                    "pre-process-backend": "va-surface-sharing",
+                    "threshold": 0.7
+                },
+                "classification-properties": {
+                    "model": "/home/pipeline-server/models/colorcls2/colorcls2.xml",
+                    "device": "GPU",
+                    "model-instance-id": "instgpu1",
+                    "inference-interval": 5,
+                    "batch-size": 8,
+                    "nireq": 2,
+                    "ie-config": "NUM_STREAMS=2",
+                    "pre-process-backend": "va-surface-sharing"
+                }
+            }
+        }
+    }
+]
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/how-to-benchmark.md b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/how-to-benchmark.md
new file mode 100644
index 000000000..52379d84a
--- /dev/null
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/how-to-benchmark.md
@@ -0,0 +1,151 @@
+# How to Run Benchmarking
+
+This guide provides step-by-step instructions for running the benchmarking script to evaluate the performance of the Smart Parking application. The script can help you determine the maximum number of concurrent video streams your system can handle while meeting a specific performance target (e.g., frames per second).
+
+## Overview of the Benchmarking Script
+
+The `benchmark_start.sh` script, located in the `metro-vision-ai-app-recipe` directory, automates the process of running performance tests on the DL Streamer Pipeline Server. It offers two primary modes of operation:
+
+*   **Fixed Stream Mode:** Runs a specified number of concurrent video processing pipelines. This mode is useful for testing a specific, known workload.
+*   **Stream-Density Mode:** Automatically determines the maximum number of streams that can be processed while maintaining a target Frames Per Second (FPS). This is ideal for capacity planning and finding the performance limits of your hardware.
+
+## Prerequisites
+
+Before running the benchmarking script, ensure you have the following:
+
+*   A successful deployment of the Smart Parking application using Helm, as described in the [How to Deploy with Helm](./how-to-deploy-with-helm.md) guide.
+*   The `jq` command-line JSON processor must be installed. You can install it on Ubuntu with:
+    ```bash
+    sudo apt-get update && sudo apt-get install -y jq
+    ```
+*   The `benchmark_start.sh` script and a payload configuration file must be available on your system.
+
+## Understanding the Payload File
+
+The benchmarking script requires a JSON payload file to configure the pipelines that will be tested. These payload files are located within each sample application's directory (e.g., `smart-parking/`). The script uses this file to specify the pipeline to run and the configuration for the video source, destination, and parameters.
+
+Here is an example of a payload file, `benchmark_gpu_payload.json`:
+
+```json
+[
+    {
+        "pipeline": "smart_parking_benchmarking",
+        "payload":{
+            "parameters": {
+                "detection-properties": {
+                    "model": "/home/pipeline-server/models/public/yolov10s/FP16/yolov10s.xml",
+                    "device": "GPU",
+                    "batch-size": 8,
+                    "model-instance-id": "instgpu0",
+                    "inference-interval": 3,
+                    "inference-region": 0,
+                    "nireq": 2,
+                    "ie-config": "NUM_STREAMS=2",
+                    "pre-process-backend": "va-surface-sharing",
+                    "threshold": 0.7
+                },
+                "classification-properties": {
+                    "model": "/home/pipeline-server/models/colorcls2/colorcls2.xml",
+                    "device": "GPU",
+                    "model-instance-id": "instgpu1",
+                    "inference-interval": 3,
+                    "batch-size": 8,
+                    "nireq": 2,
+                    "ie-config": "NUM_STREAMS=2",
+                    "pre-process-backend": "va-surface-sharing"
+                }
+            }
+        }
+    }
+]
+```
+
+*   `pipeline`: The name of the pipeline to execute (e.g., `smart_parking_benchmarking`).
+*   `payload`: An object containing the configuration for the pipeline instance.
+    *   `parameters`: Allows you to set pipeline-specific parameters, such as the `detection-device` (CPU, GPU, or NPU) and other model-related properties.
+
+## Step 1: Configure the Benchmarking Script
+
+Before running the script, you may need to adjust the `DLSPS_NODE_IP` variable within `benchmark_start.sh` if your DL Streamer Pipeline Server is not running on `localhost`.
+
+```bash
+# Edit the benchmark_start.sh script if needed
+nano benchmark_start.sh
+```
+
+Change `DLSPS_NODE_IP="localhost"` to the correct IP address of the node where the service is exposed.
+
+## Step 2: Run the Benchmarking Script
+
+The script can be run in two different modes.
+
+Navigate to the `metro-vision-ai-app-recipe` directory to run the script.
+
+```bash
+cd edge-ai-suites/metro-ai-suite/metro-vision-ai-app-recipe/
+```
+
+### Fixed Stream Mode
+
+In this mode, you specify the exact number of pipelines to run concurrently. This is useful for simulating a known workload.
+
+**To run 4 pipelines simultaneously using the CPU payload:**
+
+```bash
+./benchmark_start.sh -p smart-parking/benchmark_cpu_payload.json -n 4
+```
+
+*   `-p smart-parking/benchmark_cpu_payload.json`: Specifies the path to your payload configuration file.
+*   `-n 4`: Sets the number of concurrent pipelines to run.
+
+The script will start the 4 pipelines and print their status. You can then monitor their performance via Grafana, by using the pipeline status API endpoint, or by running the `./sample_status.sh` script from the `metro-vision-ai-app-recipe` directory to check the average FPS.
+
+### Stream-Density Mode
+
+In this mode, the script automatically finds the maximum number of streams that can run while maintaining a target FPS. This is useful for determining the capacity of your system.
+
+**To find the maximum number of streams that can achieve at least 28.5 FPS on CPU:**
+
+```bash
+./benchmark_start.sh -p smart-parking/benchmark_cpu_payload.json -t 28.5
+```
+
+*   `-p smart-parking/benchmark_cpu_payload.json`: Specifies the path to your payload configuration file.
+*   `-t 28.5`: Sets the target average FPS per stream. The default is `28.5`.
+*   `-i 60`: (Optional) Sets the monitoring interval in seconds for collecting FPS data. The default is `60`.
+
+The script will start with one stream, measure the FPS, and if the target is met, it will stop, add another stream, and repeat the process. This continues until the average FPS drops below the target. The script will then report the maximum number of streams that successfully met the performance goal.
+
+**Example Output:**
+
+```
+======================================================
+✅ FINAL RESULT: Stream-Density Benchmark Completed!
+   Maximum 4 stream(s) can achieve the target FPS of 28.5.
+======================================================
+```
+
+### How Stream Performance is Evaluated
+
+In Stream-Density Mode, the script evaluates if the system can sustain a target FPS across all concurrent streams. The process is as follows:
+
+1.  **Individual Stream Monitoring:** The script monitors each running pipeline instance (stream) independently.
+2.  **Sampling:** For the duration of the monitoring interval (e.g., 60 seconds), it samples the `avg_fps` value from each stream every 2 seconds.
+3.  **Averaging per Stream:** After the interval, it calculates the average FPS for *each stream* based on the samples collected for that specific stream.
+4.  **Validation:** The performance goal is considered met only if **every single stream's** calculated average FPS is greater than or equal to the target FPS. If even one stream falls below the target, the test fails for that number of concurrent streams.
+
+This ensures that the reported optimal stream count represents a stable configuration where all streams are performing adequately, rather than relying on a combined average that could hide underperforming streams.
+
+## Step 3: Stop the Benchmarking
+
+To stop the pipelines, run the `sample_stop.sh` script from the `metro-vision-ai-app-recipe` directory.
+
+```bash
+./sample_stop.sh
+```
+
+This script will stop all running pipelines that were initiated by the benchmark script.
+
+## Summary
+
+In this guide, you learned how to use the `benchmark_start.sh` script to run performance tests on your Smart Parking application. You can now measure performance for a fixed number of streams or automatically determine the maximum stream density your system can support.
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/index.rst b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/index.rst
index e8630ad28..8732fc27b 100644
--- a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/index.rst
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/docs/user-guide/index.rst
@@ -20,6 +20,7 @@ Using advanced technologies and pre-trained deep learning models, the applicatio
 
    how-to-customize-application
    how-to-deploy-with-helm
+   how-to-benchmark
    how-to-deploy-with-edge-orchestrator
 
 .. toctree::
diff --git a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/src/dlstreamer-pipeline-server/config.json b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/src/dlstreamer-pipeline-server/config.json
index 9b9b1b7c3..1d723b1c3 100644
--- a/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/src/dlstreamer-pipeline-server/config.json
+++ b/metro-ai-suite/metro-vision-ai-app-recipe/smart-parking/src/dlstreamer-pipeline-server/config.json
@@ -91,6 +91,31 @@
                 },
                 "auto_start": false,
                 "publish_frame": true
+            },
+                        {
+                "name": "smart_parking_benchmarking",
+                "source": "gstreamer",
+                "queue_maxsize": 50,
+                "pipeline": "multifilesrc location=/home/pipeline-server/videos/new_video_1.mp4 loop=true ! parsebin ! vah264dec ! vapostproc ! video/x-raw(memory:VAMemory) ! gvadetect name=detection ! queue ! gvaclassify name=classification ! queue ! gvapython module=/home/pipeline-server/models/colorcls2/process class=Process function=process_frame ! queue ! gvawatermark ! videoconvertscale ! video/x-raw,format=I420 ! gvafpscounter ! appsink sync=false async=false",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "detection-properties": {
+                            "element": {
+                                "name": "detection",
+                                "format": "element-properties"
+                            }
+                        },
+                        "classification-properties": {
+                            "element": {
+                                "name": "classification",
+                                "format": "element-properties"
+                             }
+                        }
+
+                    }
+                },
+                "auto_start": false
             }
         ]
     },
@@ -111,4 +136,4 @@
             }
         ]
     }
-}
+}
\ No newline at end of file