Machine Learning algorithm evaluation on advanced driver assistance

Abstract

In this research and development project, our main purpose is to study four deep learning architectures for real-time object detection of people and bicycles encountered in front of driving. We use 4 different algorithms for the same data set, and compare the mAPs obtained after training. And discuss which method is the most accurate, but also consider the time it takes to get what is suitable for what kind of scene. The project I came up with would like to be used in a driving assistance system. The system uses camera sensors to get input, and then uses algorithms to assist, so that the safety of the car is guaranteed when driving. At the same time, it can run on a lowperformance version of the machine and compare the fps of different algorithms.