Research On Pedestrian Target Tracking Method Based On 2D Human Skeleton And Structured Data

Target tracking technology is one of the key research topics in the field of computer vision,which is widely used in the field of visual tracking control such as intelligent monitoring,human-computer interaction,robots,and drones.Tracking technology faces various challenges due to the diversity of application scenarios,and the target may drift or lose in complex environments such as occlusion,overlap,rapid movement,illumination changes,and object interference.However,most of the existing tracking algorithms have not reached the mature stage in the complex environment.How to make the target achieve real-time and accurate tracking is the focus of research by researchers.Therefore,this thesis conducts in-depth research on video image preprocessing,pedestrian target feature extraction and pedestrian target continuous tracking to solve the problem that the target is difficult to retrieve after tracking loss in complex environment,the main contents are as follows:(1)Image preprocessing: The image is unclear due to the problems that light changes and image fogging during the detection of the target,resulting in a lower target recognition.Therefore,this thesis compares the illumination processing of Histogram Equalization(HE),Adaptive Histogram Equalization(AHE)and Contrast Limited Adaptive Histogram Equalization(CLAHE)on the RGB and YUV color space,the image equalization preprocessing method based on CLAHE algorithm of YUV color space Y component is designed to improve image quality and solve image illumination.The problem of both.(2)Pedestrian target feature extraction: Feature extraction is a key part of target detection and recognition.Therefore,the 2D human skeleton detection model is established through the training and tuning of the bifurcated deep convolutional neural network model to realize the pedestrian skeleton detection under various poses and partial occlusions of pedestrians,and a structured feature extraction method based on human geometric topological constraints is proposed.The coordinate information of the skeleton joint point position can be used to intercept the structural features of the various parts of the human body to realize the accurate extraction of the structural features of the human body.(3)Pedestrian goal continuous tracking: When the target occludes,overlaps,or changes in the angle of view,the target recognition degree is reduced,and the target may drift and lose.Therefore,a pedestrian target continuous tracking algorithm based on KCF and structured feature fusion is proposed.The KCF response function peak value is used to judge whether the target is lost due to occlusion,overlap and perspective change.The target recognition and relocation strategy of template autonomous updating is established by integrating structured feature matching to achieve stable and continuous tracking of target pedestrians,and to solve the problem that traditional tracking algorithms are easy to be lost and difficult to retrieve automatically in complex situations such as occlusion and multiplayer overlap.The tracking experiment is carried out in an indoor scene with dark light and large space,the results show that the proposed target tracking method can meet the fast retrieval after the target is lost and achieve the purpose of real-time tracking.