Research On Dynamic Target Tracking And Ranging Based On Binocular Vision

Binocular stereo vision,which has become a hot spot in current robot research,is an important means to track the target and obtain the depth information of the target.It is widely used in many fields such as electric power,transportation,aviation and industrial production.In these applications,tracking and ranging of moving objects is usually required.However,the performance of tracking and ranging is greatly affected by light changes and target movement.Therefore,this paper carries out research for this purpose.The main research contents and achievements are as follows:(1)This paper analyzes the mathematical principle of geometric model of camera imaging and binocular stereo vision ranging.MATLAB calibration box is used to add the image data set collected by binocular camera,and the internal and external parameters of binocular camera is calculated with Zhang's calibration algorithm.And then the input image pairs are rectified.(2)Aiming at the phenomenon that the process of binocular stereo vision target tracking is prone to instability,the kernel correlation filtering KCF algorithm is applied to track the target.Meanwhile,TLD algorithm is combined to solve the problem of tracking failure caused by scale change or target occlusion in long time tracking.The combination of the two algorithms makes the time consume of real-time tracking between TLD's and KCF's.In this paper.a target tracking method based on ROI is adopted.The ROI regions of the same size are extracted from the left and right views by using the result of target selection,and the matching feature points are selected in these regions.Such approach reduces the computation of matching greatly,which improve the performance of real-time tracking.(3)Various stereo matching algorithms based on features are evaluated,which verify that ORB algorithm has better real-time performance under the conditions of illumination and blur transformation,scale and rotation transformation,and anti-jitter.To solve the problem of higher error of ranging caused by matching error in the real-time ranging process with binocular camera,this paper proposes a stereo matching algorithm,LAAT,based on the improved ORB feature.This algorithm adopts linear brightness adjustment method to solve the problem that ORB algorithm is easily affected by illumination.At the same time,an adaptive threshold is designed to replace the fixed threshold in ORB for image feature extraction.This method has the characteristics of strong robustness and high speed.(4)The ranging system is designed.And experiments are carried out for the static and dynamic targets respectively in the system to verify the accuracy and real-time performance.Static ranging experiments are conducted on targets with different distances by binocular ranging and errors are analyzed.Dynamic ranging is conducted by moving the binocular camera and the target simultaneously,and calculates the distance between the moving target and the camera in real time.Experiments show that the designed binocular vision system can track and range the target effectively.The experimental results indicate that the overall error of static ranging of LAAT is smaller than that of SGBM when different objects are used as targets.In the experiment of dynamic ranging,the ranging error of LAAT is kept within 5%,which reflects the high ranging accuracy.In the experiments under different illumination intensity,the ranging accuracy of LAAT is less affected by the change of illumination intensity.