Moving Target Detection And Tracking Based On TOF

With the advent of the Internet + Industrial 2.0 era,the concept of smart cities is also coming out,such as intelligent transportation,driverless,security monitoring,drones,etc.are applied to real-life scenarios.Implementation techniques based on these areas are target detection and tracking in computer vision.In recent years,At the target tracking model level,algorithms based on machine learning and related filtering frameworks do improve the accuracy and speed of the model,but the target is interfered by the external environment(light,climate,etc.),target occlusion,target change speed,etc.Target tracking is still a problem under harsh conditions.In the past,most of the vision-based tracking methods were mainly for color or grayscale images.The research only stayed at the level of two-dimensional images.Some studies used modeling to transform two-dimensional into three-dimensional layers,but the process was complicated.The inherent problem of target tracking is that the RGB information of the two-dimensional plane cannot fully reflect the image information of the three-dimensional world.This paper is based on the 3D depth image adopted by TOF,which is the research of point cloud data information.Depth distance(the distance between camera and pedestrian moving target)is not affected by illumination and climate,so the characteristics based on Depth are robust.The premise of target tracking is target detection.This paper proposes an improved adaptive hybrid Gaussian model(AGMM)detection model algorithm.By adding a priori judgment of the Gaussian distribution of the background and the optimization mechanism of the update rate in the traditional mixed Gaussian model,the background model can be established quickly and accurately,and the pedestrian moving target with less shadows can be detected efficiently.Combine the innovative3 D Kalman filter(Kalman filter in 3D space)to track human moving targets in real time,and in order to eliminate the cumulative error generated by the Kalman filter in the subsequent tracking process,this paper uses the structural characteristics of the point cloud data to adopt The adaptive stratification method accurately determines thetrajectory area of the target,and finally ensures the detection and tracking of efficient pedestrian moving targets.Finally,the experimental results show that the proposed target detection and tracking algorithm can correctly detect and track human moving targets under single target,multi-target,target occlusion,and multi-directional targets,and is not affected by climate.Light and other effects.In the construction of target detection and tracking algorithm,compared with the traditional detection and tracking algorithm(HOG-based target detection algorithm,hybrid Gaussian model algorithm,Cam Shift tracking algorithm,etc.),significant results have been achieved in speed,accuracy and accuracy.