Research On Pedestrian Detection And Tracking In Complex Background

Pedestrian detection is a hot topic in recent years and developing rapidly.It has emerged a lot of good pedestrian detection methods which can meet the practical application of the detection rate and other performance requirements.Through consult literature about the pedestrian detection under complicated background,we conducted in-depth study of existing identification methods and theories.This paper studied pedestrian detection and every aspect of tracking about the existing common pedestrian detection step.Through the establishment of the global motion estimation parameters model,using fast target detection algorithm based on improved ORB features realized of pedestrian detection and tracking.Finally,using the K-means algorithm and PROSAC(modified sample consistency)algorithm to remove the mismatching points,which can accurately identify and track target pedestrians and improve accuracy.First,according to the actual situation,global motion parameters model is set up to detect and track target pedestrians accurately and timely in complex background.The parameter model can estimate the global motion accurately using more parameters,which makes the computation complexity.In this paper,based on comprehensive consideration of the calculation of the algorithm,real-time and the accuracy of the motion estimation,the camera motion parameters of the model are six affine model parameters of orthogonal projection.Secondly,in the study of detection and tracking of pedestrians in the complex background,using the fast target detections algorithm based on improved ORB feature detects and tracks pedestrians.The improved FAST feature points is detected from the target of each frame image,which can form oFAST feature point set.Using the improved BRIEF descriptor(rBRIEF)describes detected feature points.Combining with the greedy algorithm and exhaustive algorithm,we search matching point pairs.In all possible binary tests,we find out rBRIEF,which has both high variance and Non-correlation.In each image frame and the target frame,we use the improved ORB algorithm to match the characteristics point pairs of pedestrian target,which can achieve the purpose of the tracking and simulation.Finally,while matching the pedestrian target feature points in the complex background,many factors,such as camera translation,rotation and scaling,can bring interference to the final result.Using the greedy algorithm to search feature points also can produce false match.First,use K-means algorithm to filter feature point pairs in this paper.Then,use PROSAC algorithms to except outliers.We use the least squares method to obtain the parameter matrix,which can get close to the real value of parameters.Simulation results show that it can make detect and track more accurate and real-time after removing the false matches.