Research Of Real-time Semi Global Matching Algorithm

Binocular stereo vision is an important branch of computer vision,which has important application value in the field of unmanned vehicle obstacle detection owing to it can get the 3D depth information of the scene.Stereo matching is the core problem in binocular stereo vision,and stereo matching effect directly determines whether binocular stereo vision system can be put into application.However,the main contradiction of accuracy and real-time performance can not be met at the same time is always exist.The content of this paper is to improve the real-time performance of the stereo matching algorithm in the premise of ensuring the accuracy of the stereo matching algorithm.The main contents are as follows:The traditional Census transform method is over dependent on the center pixel of the transform window,when the window center pixel is disturbed by the noise,the error will be occurred.In view of the above problems,the author propose a method of Census transform based on median filtering.Compared with the traditional Census transform method,the proposed algorithm is improved by introducing a median filter window,and,instead the center value of the original transform window with the median of filter window.The improved algorithm can effectively remove the noise points in the process of median filtering,which can effectively remove the noise points,and improve the anti-noise ability of the algorithm.In traditional semi global matching algorithm,the penalty factor is difficult to set up.So,an adaptive penalty factor setting method is proposed in this paper.According to the change of the region in the scene,the algorithm adjusts the penalty factor dynamically,which can effectively reduce the matching error rate of the algorithm.For the embedded hardware platform,based on the traditional SGM algorithm,the algorithm structure is adjusted and optimized,and the SGM matching algorithm is designed for the FPGA hardware environment,The algorithm follows the Pipeline processing mechanism,and the pixel points run in the stream processing state,which can achieve good real-time performance.A Real-Time Semi Global Matching Algorithm is proposed by combining the adaptive penalty factor setting method and the structural optimization method of the SGM algorithm,Compared with the original algorithm,this algorithm reduces the requirement of hardware resources,and it has a great improvement in real-time abilities,which is suitable for the real-time embedded system with limited resources and high real-time requirement.Designing the Stereo matching algorithm IP core by using HLS.And the process and structure of stereo matching algorithm suitable for HLS synthesis are designed,and the algorithm simulation is carried out.The real-time semi global matching algorithm was realized by using FPGA hardware platform,the accuracy and real-time performance of the algorithm were compared.The experimental results show that the proposed method is superior to the traditional semi global matching algorithm in real-time and resource occupancy,and has good accuracy.It is suitable for application in the field of real-time embedded system.