The Key Technology Research Of Pedestrian Detection And Tracking

With the development of intelligent vehicles and intelligent transportation technology,the pedestrian safety problem in traffic environment is becoming more and more attention.The Histogram-Of-Gradient(HOG)descriptor has become a popular descriptor in the field of vision-based pedestrian detection.But the traditional HOG-based pedestrian detection method is difficult to meet the requirements of intelligent vehicle applications.To this end,this paper come up with a Descriptor Generation Model based descriptor searching method for pedestrian detection.This method is used to find a better descriptor than HOG.At first,this paper proposes a general model,coined as Descriptor Generation Model,which can be used to systematically construct a wide range of descriptors.In order to evaluate descriptor and search faster,this paper introduces a Pedestrian Detection Experimental Framework and a customized Genetic Algorithm.As the optimal solution,a new descriptor named Second-order Gradient of Pedestrian detection(G2P)is presented.Then this paper proposes a G2P-based pedestrian detection method.With the combination of Support Vector Machine(SVM)and a cascading detection framework,a real-time pedestrian detection method is proposed.To address the unsteadiness of detection result,this paper proposes a TLD(TrackingLearning-Detection)-based pedestrian tracking method.this paper replaces the detection descriptor and adjusts the tracking feature points to enhance the TLD's adaptability in pedestrian tracking.At last,this paper combines G2 P pedestrian detection with TLD pedestrian tracking,and proposes a pedestrian detection and tracking system.Experimental results demonstrate the advantage of this paper's methods.