High frame rate pedestrian recognition and re-storage system based on lightweight YOLOv5+Deepsort

With the continuous development of computer vision in recent years, image perception and retrieval technologies are widely used in life, and pedestrian recognition has attracted a lot of attention in recent years because of its important application prospects in the field of intelligent surveillance systems. Traditional deep learning-based pedestrian recognition research has also made great progress thanks to the iteration of algorithms, and almost all the advanced models in the field are based on deep learning technology. In this paper, we propose a lightweight “YOLOv5+Deepsort” pedestrian recognition model, which can accurately and rapidly search for the trajectory and location of pedestrians in a wide range of surveillance video networks, and then realize efficient tracking and data storage for them. It can accurately and rapidly search for pedestrian trajectories and locations in a wide range of surveillance video networks, and then realize efficient tracking and data storage. The experimental results show that the average recognition rate of a single frame in the video during batch image recognition is around 0.015 seconds, and the recognition accuracy reaches 92.7%. At the same time, the extreme lightweight of the model applies to GeForce GTX 1050 Ti, which can run smoothly for computers with poor performance called, and also has a very high recognition rate under the premise of meeting real-time.