Abstract:
In view of the problems of the existing binocular system such as large size, high cost and difficulty in transplantation, a small binocular ranging system based on the Android embedded platform was designed. The key technologies such as binocular camera calibration, image preprocessing, binocular image correction and block matching algorithm were used to realize a small binocular ranging of robot human-computer interaction system. Aiming at the problem that the parallax value caused by focclusion is unreliability, which leads to holes in the disparity map, the left-right consistency detection and fast average filtering were added to obtain the final disparity map, and the spatial three-dimensional coordinates and distance of the object relative to the binocular camera were obtained from the disparity map. The binocular vision hardware experimental platform was built to test the proposed binocular ranging scheme. The results show that the designed ranging system is small in size and low in cost. When the detected target is 400-900 mm away from the binocular camera, the detection error rate is below 1.05%, the detection time is less than 156 ms. It can give the corresponding interactive prompts for different distance ranges, which basically meets the requirements of ranging accuracy and time of human-computer interaction experience on Android platform.