Research And Implementation Of Human Detection And Tracking Algorithm In Complex Dynamic Environment

Visual Object Tracking is one of the important research directions of machine vision,image processing and pattern recognition.Due to the complexity and diversity of dynamic environment,the difficulty of research on tracking algorithm is increased.In recent years,many researchers at home and abroad have proposed some excellent algorithms for the further research of tracking algorithms,but there is still no algorithm that can be used for human tracking in any scene.The tracking algorithm is analyzed and researched on the influence factors of human deformation,illumination change and background occlusion in complex dynamic environment.The improved algorithm are proposed and validated by a large number of experiments.The contents and achievements of our research include:1.Under the dynamic environment,the advantages and disadvantages of the frame difference method for human detection.We improve and design the human detection algorithm of automatic threshold segmentation mixed model based on three frame difference method and background subtraction method,in the dynamic environment,the contour of human is extracted.By using the network model of SSD(Single Shot MultiBox Detector)detection algorithm,the ratio bounding box and training data in SSD detection structure are increased to improve the accuracy of human detection so that the follow-up tracking algorithm extracts the effective characteristics of the human body.2.On the basis of human detection results,We use Brisk algorithm to extract feature points from human regions.Combined with the feature point detection clustering method,a matching point is selected for the secondary matching feature points in the initial matching feature set,and the next frame matching library is effectively updated to complete the target tracking in the complex scene.3.Based on the research of Camshift algorithm,the HSV and YCbCr dual channel color space models are improved for the background occlusion and similar factors of the human body target and the rapid movement of the human body,which can not be detected within the search range.The use of iterative method to update the target center of thehuman body to carry out continuous correction,update the human target model to achieve adaptive detection results,the Kalman filter and the Markov model are used to predict the position of the target at the next time,and the search area centered on the position is used to effectively reduce the influence of the detector detection range and the interference factor and improve the tracking accuracy in the complex dynamic environment.Aiming at the complex dynamic background research and improvement of the tracking algorithm,and in-depth research of tracking algorithm based on deep learning for complex scenes and algorithm performance.Through the data validation in many different scenarios,the improved algorithm can effectively reduce the influence of human body deformation,illumination change and background occlusion in the complex scene,realize the accurate and real-time tracking of the human,which makes the tracking algorithm more robust.