Activity Recognition And Traffic Flow Prediction In Surveillance Video

With the development of big data and artificial intelligence,intelligent video surveillance technology has provided a great guarantee for public safety.As an important part of intelligent video surveillance system,activity recognition technology is a research hotspot in computer vision,and it is also one of the challenging tasks.Vehicle detection technology has attracted much attention as a basic technology in intelligent transportation.This thesis focuses on activity recognition and vehicle detection detection in surveillance video.For activity recognition and detection,this paper proposes a key pose recognition method that combines motion foreground detection,and adds a behavior recognition module based on temporal domain information,using the spatiotemporal information of the video to improve the recognition effect.This method achieved the first place in the Riding、Activity_Carrying and Transport_HeavyCarry recognition of TRECVID ActEV 2018 evaluation,and the fourth place in Pull.The evaluation results show that our method can better extract discriminative features in key poses.In ActEV-PC 2019 evaluation,we applied the event detection method based on spatiotemporal domain information to the behavior of Pull and Transport Heavy Carry,and achieved better evaluation results than in 2018.We apply Faster R-CNN to vehicle detection tasks in crowded scenarios,and add FPN module to improve the accuracy of vehicle detection,and compare the detection effect of YOLO V3.