Research On Stereo Matching Algorithm In Binocular Stereo Vision

Stereo vision is one of the most extensive areas of computer vision research and is widely used in areas such as autopilot,3D reconstruction,and obstacle detection.Stereo vision obtains pictures from different perspectives of the same scene through the left and right cameras,and then obtains the depth information of the scene through calculation.Stereo matching is the process of calculating the horizontal deviation of corresponding points from the left and right pictures.However,due to the effects of illumination,shielding,and texture,stereo matching itself has great challenges and becomes the most core and difficult part of the stereo vision system.In practical applications,the time-consuming of the stereo matching algorithm is the key to the real-time system under the condition of ensuring the accuracy.For the stereo matching algorithm,the semi-global algorithm has been widely used because of the balance of precision and time consumption of the algorithm itself.Based on the semi-global algorithm,we have improved and optimized it.The specific results are as follows:1)A gradient-based semi-global algorithm is proposed.Census transform is a method of calculating the matching cost.Its robustness to light changes and good matching results make it widely used.In order to improve the efficiency of the algorithm and reduce the computational cost of the algorithm,only eight directions of pixels are calculated when calculating the cost,and a new cost model is constructed through gradient changes.The cost value is obtained by solving the Hamming distance of the cost model.The algorithm greatly reduces the amount of calculation while ensuring the accuracy is basically unchanged,and it is easy to perform parallel processing and improve the efficiency.2)A semi-global algorithm based on video sequences is proposed.Most of the previous algoritlums deal with still pictures,but for stereoscopic videos,inter-frame correlation optimization algorithms can be used.A dynamic disparity range and a time direction path are proposed by using the inter-frame information.The former uses the disparity value of the previous frame to reduce the search range of the current frame and handle the abnormal point.The latter uses inter-frame information to construct a time-oriented path during cost accumulation.The dynamic disparity range significantly reduces the computational complexity of the algorithm with less loss of accuracy.On the other hand,the time direction path increases the accuracy of the algorithm when the consumption calculation is not large,especially when the computational resources of the algorithm are limited,and the number of paths for accumulating costs is required.