Parking Space Detection

Table of Contents

0. Purpose
1. Setup
2. Personal Project Description
3. Imlpementation status

0 Purpose

The purpose of this library is to setup a remotely accessible camera that scans a static location for available parking spots and informs the user about it (i.e., you should be able to trigger a camera via phone/email and receive a message/email about whether a given parking area has a free spot available or not)

1 Setup

1a Conda environment

Install the corresponding conda environment via
conda env create -f environment.yml
(and don't forget to activate it via conda activate cam)

1b Testing

• Unit tests
  Run unit tests via
  python -m unittest
  Note: if you are using PyCharm, I would recommend to setup the testing environment as is shown here
• Integration tests
  TBD

1c Formatting

Run black code formatting from the project root directory via
black --line-length 120 .
Note: if you are using PyCharm, I would recommend to setup black as is shown here as default code formatter with the additional argument line-length 120 triggered during every save operation.
Black version used in the code is 21.12b0.

2 Personal Project Description

This Repo is built for the purpose of remote parking space monitoring and based on 2 main parts (and 4 resulting features)

1. Image acquisition part:

• Feature A > Camera setup: Building a camera setup that allows remote access (e.g., via smartphone)
• Feature B > Server setup: Sending the acquired image to a (remote) server

2. Computer Vision part:

• Feature C > Parking space detection: Detection of free parking space based on acquired image
• Feature D > Information sharing: Sharing of results (e.g., just sharing raw image and/or Yes-No result of CV algorithm - i.e., yes there is a free parking spot)

Feature A > Camera setup:

Requirements:

• Needs to be accessible remotely (e.g., via smartphone and/or computer) - e.g., send an e-mail to server -> image acquisition is triggered Initial idea of solution:
• Hook up camera to RaspberryPi which is connected to WiFi
• Triggering TBD
• Image needs to be uploaded to server (can be just simple sharing to Laptop in beginning)

Feature B > Server setup:

• TBD -> Can be done on local machine in beginning; AWS EC2 instance later

Feature C > Parking space detection:

• Probably some simple CNN (e.g., from detectron2) - but should be able to fine-tune based on specific scene of interest (which will be static - i.e., always the same)

Feature D > Information sharing:

• Once results are available, should be shared via e-mail - (can be just sharing of the raw image in beginning)

3 Implementation status

	Level 1		Level 2		Level 3
Feature	Description	Status	Description	Status	Description	Status
A > Camera setup	Trigger laptop/smartphone camera from device in same WiFi (e.g., via SSH)	-	Trigger laptop/smartphone camera from device outside of network (e.g., via VPN & SSH?)	-	Standalone camera module connected to Wifi (e.g., RaspberryPi)	-
B > Server setup	tbd: will likely do this last and run computations locally first	-	-	-	-	-
C > Parking space detection	Yolov4-tiny to infer number of cars in image (and if less than given number -> assume free spot)	✅	TBD	-	Fine-tuned CNN predicting whether free parking spot is available (i.e., yes/no vs n_cars)	-
D > Information sharing	Programmatically send plain-text e-mail to specified address	✅	Send attachments (e.g., image)	-	n/a	-
E > Forecasting	Take multiple images of parking lot over extended period of time and detect patterns (i.e., forecast expected n_cars in parking lot)	-	n/a	-	n/a	-