Research On The Synthetic Depth Map Algorithms Of Two-viewpoints

In order to imitate the ability of HVS (Human Vision System) to identify the space, the research on computer stereovision system raises its curtain, which analyzes and processes the information of scene images acquired by means of the technology of the video camera and computer device to get the geometric information of physical object, such as shape, size, motion and so forth. Two-viewpoints stereovision is a fundamental measure in science and technology of machine vision. Although there are still some technical difficulties to overcome, it has very broad application prospects in many diverse fields such as military applications, robot vision navigation,3D movies, monitoring system and medical device.This article mainly focused on the research on the synthetic depth map algorithms of two-viewpoints, which adopted two cameras from different positions for collecting to the picture information of the same scene in order to build three-dimensional imaging system. Then it used a new synthetic depth map algorithm proposed to obtain a dense disparity map. According to the given camera parameter and the geometric relation between disparity map and depth map, it has finally synthesized a depth map of the target scene. To proactively satisfy the demand of the engineering, the proposed algorithm has completed the further work of real-time analysis and optimization.The research mainly included three aspects:(1) A new two-viewpoints camera calibration method was proposed in this section by improving the encoding method and changing the crossover and mutation probabilities. The improvement of the encoding scheme was able to adjust the search space adaptively. The changing possibility of the crossover and mutation was able to make the population to develop in diversity and to avoid premature convergence. The new two-viewpoints camera calibration method proposed has provided not only a good solution to the nonlinear optimization problem with multiple parameters and complex optimization function, but also much higher accuracy. At the same time, the operation rate of algorithm has improved 47 percent.(2) A new synthetic depth map algorithm was proposed, which used the revised adaptive support-weight cost function and constructed a new global energy function. Then the improved bidirectional dynamic programming approach was presented for looking for route. Finally after mismatches were removed by use of the simple filtering method and three reliability rules in order to obtain a dense depth map. The experiment results reveal that the algorithm is better effective both in decreasing the overall matching error rates and reducing apparent defective stripes in comparison to the other three algorithms.(3) The new algorithms proposed had completed the further work of real-time analysis and optimization. The new two-viewpoints camera calibration method builded the parallel programming model based on OpenMP in order to achieve timeliness improvement. Furthermore, the new synthetic depth map algorithm programmed in the OpenCV library and provided optimized implementations based on IPP technology. The experiment results show that the parallel operation of the new two-viewpoints camera calibration method only costs 34 percent of serial operation, Meanwhile, the run rate of the new disparity map acquisition algorithm optimized costs 46 percent of its functioning before optimization. According to the real-time analysis and improvement of algorithms, their performance has significantly increased and satisfied the actual demand.