Study Of Vision Based Pedestrian Detection Algorithm And System Design

With the rapid growth of vehicles running on the road in the last decade, road safety has become a worldwide issue and pedestrians are the most vulnerable group in traffic accidents. Pedestrian detection system, as a part of the vehicle’s advanced driver assistance systems, is developed from the perspective of vehicle active defense to protect road pedestrians. With the help of pedestrian detection system installed in a vehicle, pedestrians involved traffic accidents can be can reduced or prevented and the driving process can be safer.Firstly, after researching the recently proposed pedestrian detection algorithms, an algorithm based on integral channel features and cascade classifier was proposed in this paper. The integral Channel Features were used as pedestrian representing features; the second-order decision tree was used as weak classifier; the Adaboost algorithm and Soft Cascade algorithm were used to training the cascade classifier; a sliding window based pedestrian detection algorithm was implemented to realize full image pedestrian detection and the classifier pyramids and crosstalk cascade algorithms were used to improve detection rate.Secondly, the DaVinci embedded system based on TMS320DM6467T processors was used to realize the pedestrian detection system. The main work in this part included the design of hardware and software of DaVinci system and the migration and realization of pedestrian detection algorithm on DaVinci system.In the end, we firstly trained a cascade classifier with target detection rate at95%and target false alarm rate at10-6via PC. Then the cascade classifier data and the pedestrian detection algorithms were migrated to the DaVinci embedded systems to achieve a practical pedestrian detection system.