Research On Pedestrian Tracking Method In The Similarity Between Moving Object And Background

Pedestrian tracking is a research hotspot in the field of computer vision,which has a broad application prospect in intelligent monitoring,automatic driving and human-computer interaction,among which the low tracking accuracy caused by the similarity of target and background is a focus of the current tracking algorithm research.In recent years,pedestrian tracking technology has been studied at home and abroad,and some breakthroughs have been made,but the actual effect still can not meet the user's requirements for high accuracy and real-time of the system.This thesis studies the problem of pedestrian tracking in similar background,and finds that one of the important factors affecting pedestrian tracking effect is that similar background interferes with the feature extraction and matching of the target,resulting in the reduction of tracking accuracy of the algorithm to the target and even the loss of the target,among which the most common one is the color similarity or texture information similarity between the target and the background.To solve this problem,this thesis proposes an adaptive method to assign feature weights: in the pedestrian tracking algorithm,the color feature represents the color of the target,and the LBP feature represents the texture of the target.Through adaptive fusion,the complementary advantages of the two features can be realized.According to the color similarity and texture similarity of the target and the background,the feature weights are automatically assigned when the pedestrian target and the background When the color similarity of the scene increases,the weight of the color feature will be automatically reduced,and the weight of other features will be increased to track;similarly,when the similarity of the texture information of the pedestrian target and the background increases,the weight of the texture feature will be automatically reduced,and the weight of other features will be increased to track,so as to enhance the robustness and accuracy of the algorithm in similar background.The specific research contents are as follows:(1)Segmentation of the target and background.By combining Gaussian and morphological processing,a clear and complete region segmentation is achieved for the target and background in the image.(2)The similarity calculation of the regional features of the target and the background.According to the result of region segmentation of target and background,the correlation similarity calculation method is selected to calculate the color similarity and texture similarity between the target region and the background region.(3)Pedestrian tracking method based on adaptive feature fusion.According to the color similarity and texture similarity to determine the feature fusion strategy,using particle filter tracking algorithm for pedestrian tracking,and the tracking performance of the algorithm is compared to verify the feasibility of the algorithm.Experimental results show that compared with other algorithms,the algorithm in this thesis can achieve better tracking effect in similar scenes,effectively solve the problem of tracking accuracy reduction caused by similar background,and has stronger robustness.