Design And Implementation Of Real-tine Binocular Stereo Vision System

The principle of binocular stereo vision system is to simulate the human eye, using standard stereo matching algorithm to analyze the depth information of the object in the image, so as to achieve the object of study of three-dimensional reconstruction, distance measurement and other functions. In recent years, the majority of stereo vision analysis is in the optimization of previous algorithms, and proposed a variety of stereo matching algorithm. Now comparing the performance of the new stereo matching algorithm has been able to match to the expected scene depth. But most of the algorithms are very large proportion of software to achieve, it is difficult to meet the needs of real-time. With the development of image technology, the hardware architecture of the binocular stereo vision system has been widely studied, but the hardware implementation in the accuracy analysis of resource occupation and contradictions, in general, complex algorithms require considerable hardware resources, some of the new algorithm by hardware realization method is not suitable for.Based on the above reasons, this paper based on the existing algorithm, resulting in a hardware chip through the algorithm of stereo matching, the specific scheme is in the integration of absolute pixel difference and "SAD", "Census" two cost algorithm for processing cost after initialization, through the use of dynamic programming method to optimize the "DP" the scan line, to ensure the data fluctuation depth map tends to smooth and filter the more noise, then with the background delete, median filter to optimize the depth map of a variety of measures, which can greatly enhance the parallax effect, and can ensure the processing speed.At the same time, this algorithm builds the hardware circuit of a high performance, automatic focusing and connecting AcuteLogic HD images using spartan6-150 T development tools Avnet camera equipment to implement PE1005 S. Through the test, this program only consumes the entire FPGA 1/4 of the hardware resources. According to the experimental data obtained in this study can be completed immediately, the highest resolution 720 P frames than the depth of figure 60, this optimization is to compare the performance of the already existing hardware solution, while the depth map results with other ways compared to a big increase, which can know it has to take into account the two characteristics. Are occupied less resource and analysis precision is high. Although there is still room for improvement, but the existing situation has been able to achieve as a passenger flow statistics, gesture recognition, such as the needs of the traditional application of product development.This paper first describes the development situation of the binocular stereo vision at present, and then introduces the principle and application of several algorithms, deduce the design principle of this subject, then the hardware system of the program detailed, finally through the data to conduct a comprehensive analysis, it is proved that the system is advanced and reliable.