Research And Application Of Multi-scale Fast Pedestrian Detection Algorithm Based On YOLO

As the monitoring scene continues to increase,pedestrian detection has also developed rapidly.Pedestrian tracking,behavioral analysis and other techniques based on pedestrian detection support have become commonplace in people's lives.The continuous advancement of pedestrian detection technology can fundamentally drive the application and popularization of upper-layer applications.In the actual pedestrian detection system,not only the accuracy of the target detection but also the real-time nature of the detection should be considered.This requires us to ensure the accuracy of the detection and the efficiency of the detection when designing the pedestrian detection algorithm.This paper combines pedestrian detection theory with practical application,and implements multi-scale fast pedestrian detection algorithm and real-time pedestrian detection system based on Tiny-YOLO.Based on the characteristics of pedestrian detection,this paper finds that Tiny-YOLO has some problems in the field of pedestrian detection and makes corresponding improvements on this basis.In view of the problems that Tiny-YOLO has in the field of pedestrian detection,the improvements in this paper are:(1)In the detection of the single target of the pedestrian,the loss function is improved to make the model better for pedestrian detection;(2)For the missed detection and misdetection,the feature extraction capability is increased by designing the network structure;(3)For the problem of inaccurate detection of small targets,the feature pyramid and multi-scale feature prediction are used to fuse different feature graph semantics,which can easily find small target pedestrians and improve detection accuracy.This paper also verified the effectiveness of the above improvements through experiments.The improved algorithm improves the real-time detection speed of pedestrians,and also improves the accuracy of pedestrian detection,which basically reaches the level of application landing.Finally,a real-time pedestrian detection system is combined with the improved pedestrian detection algorithm to further verify the availability and practicability of the algorithm.