Deep Learning Model on Embedded Board for Vehicle Detection and Vehicle Tracking
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This paper proposes a deep learning model to detect and track the vehicle on an embedded device such as Odroid, Orange Pi, etc. This system includes two main parts: vehicle detection and vehicle tracking. Since deep learning has achieved high accuracy over the classical image processing method, object detectors can detect vehicles in the street and highway. It can be normal to run the computer detection program with graphic processor unit (GPU) support, but it is challenging to run it on the embedded board with no GPU support and low central processing unit (CPU) performance. This paper focuses on balancing edge-computing-based deep learning object detection's accuracy and performance using additional techniques such as quantization, edge TPU, and multiple threads. SSDLite with MobileNet backbone is chosen due to its lighter than other networks but still obtain good performance compare with Yolo.
Abstract
This paper proposes a deep learning model to detect and track the vehicle on an embedded device such as Odroid, Orange Pi, etc. This system includes two main parts: vehicle detection and vehicle tracking. Since deep learning has achieved high accuracy over the classical image processing method, object detectors can detect vehicles in the street and highway. It can be normal to run the computer detection program with graphic processor unit (GPU) support, but it is challenging to run it on the embedded board with no GPU support and low central processing unit (CPU) performance. This paper focuses on balancing edge-computing-based deep learning object detection's accuracy and performance using additional techniques such as quantization, edge TPU, and multiple threads. SSDLite with MobileNet backbone is chosen due to its lighter than other networks but still obtain good performance compare with Yolo.