This is the repo accompanying my Medium article "Edge AI - Computer vision on the edge", Part 1, Part 2.
Whenever network connectivity is not available, or network latency is insufficient, or there is a need for local processing due to regulatory, privacy and security concerns, or, indeed, a combination of all of these aspects is given, today’s standard cloud-based AI approach isn’t an option.
Edge AI, i.e., AI computations performed locally on the “edge” with little or no network connectivity using locally obtained data represents a viable alternative under these circumstances subject to the computation and data processing limitations of the edge device under consideration.
In the Medium blog posts, I explain how to use the open source Intel OpenVINO library together with the Intel Neural Compute Stick 2 for edgde AI in the form of object detection on a commodity edge device, the Raspberry Pi.
The corresponding object detection demo application, which is produced in the course of the two blog posts, using YOLOv3-tiny is shown below (input video source: CAMPUS dataset for multi-view object tracking - Y. Xu, X. Liu, L. Qin, and S.-C. Zhu, “Cross-view People Tracking by Scene-centered Spatio-temporal Parsing”, AAAI Conference on Artificial Intelligence (AAAI), 2017).
The hardware setup looks as follows:
- Raspberry Pi 3 Model B+ (Raspberry Pi 4 should be just fine as well)
- Raspberry Pi on-board camera module
- Intel Neural Compute Stick 2
The software side of things consists of:
- Intel OpenVINO toolkit for Raspbian OS to be installed on the Raspberry Pi. The demos were generated with version: 2020.1.023
- Intel OpenVINO toolkit for
,
or
for your main machine
- At least Python 3.5.6
I recommend to install FFMPEG on your Raspberry Pi to make it easier to find a working file extension and output video codec combination.
The custom-implementation of object detection applications using this hardware and software setup alongside the code in this repo is described in detail in Part 2 of my Medium article.