An End-to-End Traffic Vision and Counting System Using Computer Vision and Machine Learning: The Challenges in Real-Time Processing

摘要

The goal of this research is to design and develop an end-to-end system based on computer vision and machine learning to monitor, count, and manage traffic. The end goal of this study is to make our urban transportation safer for our people, especially for pedestrians and bicyclists, who are the most vulnerable components of traffic collisions. Several methods have been proposed for traffic vision, particularly for pedestrian recognition. However, when we want to implement it in real-time in the scale of a large city like Los Angles, and on live video streams captured by regular traffic cameras, we have to deal with many challenges. This paper introduces the main challenges in traffic vision in practice, and proposes an effective end-to-end system for traffic vision, detection, tracking, and counting to address the challenges.