Design and Implementation of a Lightweight Object Detection System for Resource-Constrained Edge Environments

TL;DR

This paper presents a lightweight, privacy-preserving object detection system optimized for resource-constrained edge devices, using model compression techniques to enable real-time detection of people and vehicles on microcontrollers.

Contribution

It introduces a novel combination of model compression methods applied to YOLOv5 for efficient embedded object detection on STM32H7 microcontrollers.

Findings

Achieved significant reduction in model size and computation.

Enabled real-time object detection on low-power microcontrollers.

Maintained acceptable detection accuracy despite compression.

Abstract

This project aims to develop a system to run the object detection model under low power consumption conditions. The detection scene is set as an outdoor traveling scene, and the detection categories include people and vehicles. In this system, users data does not need to be uploaded to the cloud, which is suitable for use in environments with portable needs and strict requirements for data privacy. The MCU device used in this system is STM32H7, which has better performance among low power devices. The YOLOv5 system is selected to train the object detection model. To overcome the resource limitation of the embedded devices, this project uses several model compression techniques such as pruned, quantization, and distillation, which could improve the performance and efficiency of the detection model. Through these processes, the model s computation and the quantity of model parameters could be reduced, in order to run computer vision models on micro-controller devices for the development of embedded vision applications.