3D Detection And Tracking Of Dynamic Objects In A Scene Based On Binocular Camera

Dynamic object detection technology has been widely used in various fields of life.Under the premise of studying the development status and application of dynamic object detection technology,this paper proposes the use of the traditional dynamic object detection,which is sensitive to noise and light changes,and the object's reciprocating motion has the problem of “ghost” false detection and inability to detect depth information.Binocular camera performs the method of three-dimensional detection of dynamic objects,and builds experimental devices and experimental scenes,analyzes and studies key technologies,and writes code for algorithm implementation.An improved adaptive Gaussian background modeling algorithm is proposed.The depth map obtained by the binocular camera is used to model the three-dimensional background,so as to detect the dynamic objects in the scene.The algorithm detects the result accurately,is not affected by the light,does not appear "ghosting" phenomenon,can get the depth information of the moving object.The work of this paper mainly includes the following aspects:1.The imaging principle of binocular stereo vision system is studied.The calibration process of binocular camera,stereo rectifying method,image matching,disparity map calculation and depth reduction are analyzed.Based on the binocular vision imaging system,the model of the line array binocular system is deduced and analyzed,and the imaging principle of the line array model is analyzed.2.The detection technology of dynamic objects is studied,including the principle and algorithm implementation of inter-frame difference method,background subtraction method and optical flow method.The detection effects are compared and their respective advantages and existing problems are analyzed.In particular,the background subtraction method based on adaptive mixture Gaussian background modeling is studied.The results of the algorithm for dynamic object detection are tested.The "ghost" problem of the algorithm and the false detection caused by light changes are proposed.The mixed Gaussian background modeling solves the influence of light changes on the detection results,and removes the modeling errors caused by "ghosting" by changing the learning rate of background modeling.3.Design a line array binocular imaging system and establish a real-time monitoring system for the scene.The system is not affected by the light,which can quickly detect whether an object enters or leaves the monitoring area,and can obtain rough depth information of the object entering or leaving the monitoring area.4.Build up the binocular vision imaging system,write the implementation code of the binocular vision imaging system,and use the binocular system to detect the depth information of the object.Combining the area binocular vision system with the improved Adaptive Gaussian Mixture background modeling algorithm,the background image of the depth image is subtracted,the dynamic objects in the scene are detected and the depth information of the dynamic object is obtained.It is proved by experiments that the algorithm can detect under the condition of light change,and can eliminate the false detection of "ghost",and can obtain the accurate depth information of moving objects.It is the real object detection and tracking in 3-D space.The binocular vision system combined with the improved adaptive hybrid Gaussian background modeling method is used to detect and track the cell entrance flow,and realize the flow statistics of the inbound and outbound cells.The statistical analysis of the measurement accuracy of the system at different precisions was carried out.