The Research Of Object Detection And Location Based On Hybrid Stereoscopic Vision System

Many applications in computer vision,such as video surveillance,virtual reality,and vision-based motion control,have benefited from the expansion of vision.Compared with the traditional cameras,panoramic vision has a horizontal 360-degree view,which has become a hot object in the field of computer vision.However,since the low and uneven resolution of the panoramic,it can only provide rough information of the surrounding environment.The system adds a high-resolution traditional perspective camera to obtain detailed information of the region of interest.Due to the different imaging of the two cameras in the hybrid stereo vision,the difficulty of target detection and positioning is greatly improved,the stereoscopic vision of the different types of cameras is behind of the same type cameras.In order to solve the above problems,this paper proposed the stereo calibration algorithm for hybrid stereo vision,implemented the target collaborative detection and three-dimensional positioning in hybrid stereoscopic vision system.Firstly,the composition and principle of the hybrid stereo vision system were analyzed.And the imaging model of the single-view panoramic camera and the traditional perspective camera were studied.The related theories of single-view panoramic vision were discussed,including imaging principle,equivalent projection model,cylindrical and expansion algorithms.Based on the single-view panoramic vision imaging process,it can be derived the equivalent spherical projection model.The equivalent spherical unified model is suitable for camera models satisfying the single-view constraint.And it has simple calculation process and few model parameters.The advantages of the traditional perspective camera also meet the single viewpoint constraint,as a special case of the spherical unified model.Secondly,the target collaborative detection method in hybrid stereo vision system was discussed.Due to the wide monitoring range of the catadioptric panoramic camera in the hybrid stereo vision system,the target can be found in the first time.The targets in the panoramic image are not easy to miss and lose.In view of the low resolution and distortion of the panoramic image,the targets were detected in the panoramic cylindrical image.Then we calculated the azimuth angle of the selected target.The high-resolution perspective camera captured detailed information obtained and realized the collaborative detection in the hybrid stereo vision system with the combination traditional and convolutional neural networks detection.Aiming at the vertical hybrid stereoscopic vision system,the stereo calibration and correction algorithm based on the spherical unified model was proposed.The structure and positioning principle of the hybrid stereo vision were analyzed in details,including the calibration of the panoramic and the perspective camera,the hybrid stereoscopic correction and 3D positioning.Based on the spherical unified model,the hybrid stereoscopic vision was corrected by the stereo image pair and the disparity map was obtained by stereo matching.The 3D localization of the target can be realized according to the triangulation and parallax principle.Finally,hybrid stereo vision was applied to the mobile robot.The flexibility of the mobile robot was used to solve the direction change of the perspective camera.At the same time,the human-computer interaction interface was designed,including image processing and robot motion control.The experimental results were analyzed and verified the effectiveness of collaborative detection and localization of the hybrid stereoscopic vision.