Fast Object Detection on the Road

Abstract

Autonomous vehicles using Artificial Intelligence (AI) technologies requires various sensors such as radars, lidar, ultrasonic, and etc. to mimic the human visual perception in monitoring the road condition. Wide-angle camera is also often adopted for better coverage of view. Those sensors generates massive amount of data that could be processed with the cloud computing through the wireless communication. However, the cloud computing may not be a feasible solution, such as for real-time detection systems. In this work, we examine the implementation of the deep-learning (DL) real-time object detection models on the edge devices that is connected to the wide-angle camera. This visual system can achieve real-time object detection with a latency of less than 0.2 ms. The DL model also help to mitigate the distortion that is introduced by the wide-angle camera. Such a detection system will be able to warn the user of his or her surrounding road conditions.