Research On Depth Perception Based On The Binocular Stereo Vision

Computer vision is aimed at using computer to simulate the human visual system.As a classical problem in computer vision area, binocular stereo vision can imitate thecharacteristics of human eyes, and restore the depth information by calculating a disparitymap with a pair of images. Compared with the active ranging method, binocular stereovision has a lot of advantages such as a simple structure, large measurement range andaccurate measurement results. That makes it an inevitable direction for the robotnavigation, medical imaging and virtual reality.Currently, the distance sensors are widely used in the obstacle avoidance and3Dmodeling for unmanned rotorcraft, which cost a lot and have low robustness. This paperresearches depth perception based on the binocular stereo vision technology, and appliesit to the unmanned rotorcraft.The binocular stereo vision technology includes the research on calibration,distortion, correction, stereo matching, depth calculation and so on. This paper comparesthe mainstream camera calibration method, and chooses the Zhang zhengyou methodwhich is more sophisticated and simple. Brown and Fryer’s lens distortion modeling isalso used to correct the distortion. Combined with the parameters of camera, this paperuses Bouguet algorithm to achieve stereo calibration.Stereo matching is the key of binocular stereo vision, and it can be divided into localand global genres. Compared to the global algorithm, the local algorithm needs lessinformation and lower computational complexity, but also a worse performance.Considering the actual requirements of UAV, this paper firstly uses local algorithm toquickly obtain the primary matching results which can provide real-time data for UAV’sdecision, then it uses global algorithm calculating a fine matches based on the primarymatching data which eventually give the correction information for UAV. Specifically, thesegmentation-based adaptive support algorithm is selected in the local stage, and the dynamic programming is used in the global stage. The comparative experiments showthat the improved algorithm achieves a more satisfactory performance.Finally, this paper takes the unmanned rotorcraft as application platform, anddesigns some experiments to computer the target depth. The experimental results showthat this binocular stereo vision has high accuracy of measurement, and is suited to theUAV.