Research On Pedestrian Detection And Tracking In Driver Assistance System Based On Deep Learning

With the development of artificial intelligence,producing smart cars equipped with driver assistance systems has become the future of automotive industry.The driver assistance system helps drivers to better understand road conditions,identify vehicles and pedestrians,and provide them with warning before potential dangers to ensure driving safety.Pedestrian detection and tracking technology plays an indispensable role in the driver assistance system,so it is important to study it.Currently,deep learning methods have gained world-renowned achievement in the field of computer vision.The target of this paper is to use the deep learning methods to realize the pedestrian detection and tracking algorithm which can fulfil the requirements in terms of accuracy and real-timeness for driver assistance system.The main work of this paper focuses on two aspects:pedestrian detection and pedestrian tracking.The main content and innovations are as follows:For pedestrian detection,a new method based on multi-fusion features is proposed.The method fuses features between high-dimensional and low-dimensional and features in different types by adopting the feature pyramid structure and an additional feature extraction branch.The proposed convolutional neural network(CNN)improves the ability of small target detection and better distinguishes between pedestrians and backgrounds.At the same time,the method predicts fusion features of different sizes separately,which helps detect pedestrians of different sizes with more accuracy.For pedestrian tracking,a multi-target pedestrian tracking method based on CNN feature association is implemented.Firstly,the Kalman filter algorithm combined with the Hungarian algorithm is used to realize a high-speed tracking framework based on pedestrian detection results.Then,an efficient CNN is used to extract features of detection and tracking results,which are then used to measure the degree of matching between them.Finally,the temporal sequences composed of the CNN features from the tracking result and the cascade matching method are used to solve the tracking discontinuity caused by occlusion.Eventually,we realize an accurate and continuous pedestrian tracking.The pedestrian detection and tracking algorithms proposed in this paper have achieved great accuracy and real-time performance on the public dataset,which meets the requirements of the driver assistance system,and also contributes to the research work in related fields.