Research On Target Pedestrian Detection And Tracking Algorithm In Video

Target detection and tracking technology in video is a major research hotspot in the field of computer vision,has favorable development prospects in applications such as traffic management,human-computer interaction,and intelligent monitoring.Although at present domestic and foreign research scholars have achieved certain research results in the field of pedestrian detection and tracking,and continue to make progress in theory and innovation,there are still many problems in practical applications.Such as the pedestrian's multi-pose problem,the continuous change of light intensity and the situation of occlusion cause the result of detection and tracking to be very different from the expected.In order to solve these problems,this paper made detailed comparative analysis experiments and research,and proposed improved target pedestrian detection and tracking algorithm,which improves the effect of pedestrian detection and tracking in practical applications.In terms of target pedestrian detection,this paper proposes a multi-feature fusion pedestrian detection algorithm and conducts experiments on the INRIA public data set.Compared with the traditional single-feature pedestrian detection algorithm,it improves the final pedestrian detection effect.The specific method is as follows:(1)Pedestrian detection algorithm based on multi-feature fusion.The basic features of the video image,HOG,SIFT,LBP and Harr features,etc.were analyzed.HOG features are used to describe the characteristics of local direction gradient information and SIFT features have the characteristics of stable local features.The HOG and SIFT features are serially fused and dimensionality-reduced.A pedestrian detection algorithm with HOG + SIFT feature fusion is introduced.(2)Experimental verification was conducted on the public data set of INRIA pedestrians.The missing and false detection rates of single-feature and double-feature detection algorithms were counted,and the ROC curve was compared.Based on this,three basic feature fusion pedestrian detection algorithms are theoretically analyzed.The experimental results show that the use of multi-feature fusion detection algorithm can improve the detection accuracy rate and effect under the change of pedestrian posture and complex background.In the process of tracking pedestrians,problems such as the interference of the background color of the environment,the change of light intensity,and the obstruction of pedestrians by obstacles have a greater impact on the tracking performance.In order to solve these problems,the following methods were adopted to improve the tracking effect:(1)The principle of the traditional Camshift(Continuously Adaptive Mean-SHIFT)tracking algorithm is elaborated.When the background color is similar to the target pedestrian and the lighting changes are obvious or the occlusion phenomenon is encountered,tracking will fail.In order to solve this problem,an algorithm based on Camshift for fusing LBP features with the H component of the HSV model is proposed,and simulation experiments are performed on the OTB100 data set.The experimental results show that the modified Camshift algorithm improves the pedestrian tracking effect under illumination changes.(2)The Kalman filter idea is introduced based on the modified Camshift algorithm.When the target is blocked,the prediction mechanism of the Kalman filter is used to predict the location of the target pedestrian in the next frame.In the occlusion experiment,Camshift combined with Kalman filter pedestrian tracking algorithm accurately tracked the location of pedestrians after occlusion,which improved the effect of target pedestrian tracking in the occlusion phenomenon.The modified algorithm proposed above can improve the problems that occur when the target pedestrian is detected and tracked,and can be used in complex environments such as pedestrian posture changes,similar background color interference,and occlusion phenomena.This allows pedestrian detection and tracking systems to be better applied to the fields of traffic monitoring,urban security,assisted driving,and human-computer interaction.It is of great significance for the practical application of pedestrian detection and tracking systems into society.