Research On Distinguished Target Ranging Technique For Infrared Binocular Stereo Vision

Binocular stereo vision is an important branch of computer vision. With the advance of infrared sensors, application prospect of infrared stereo vision has been greatly expanded in military and industry. In this paper, we have deeply researched the calibration, epipolar rectification and stereo matching and designed a distance measuring system based on infrared binocular stereo vision.When calibrating infrared cameras, conventional checkerboard is inoperative because the checkerboard has no mutative temperatures. A special transmission checkerboard, which is based on active infrared radiation, has been manufactured, introducing the classical planar pattern calibration method into infrared cameras calibration.In order to solve the problem that epipolar constraint is not satisfied for image pairs generated by unconstraint infrared stereo rig, an improved epipolar rectification method has been presented. Modified K-nearest neighbor is utilized to eliminate outliers, replacing conventional RANSAC. Then the rectification matrices are obtained by particle swarm optimization, utilizing the idea of chaos optimization and hybridization in genetic algorithm. Experiments show that our rectification method is adaptive and robust and has better accuracy.Due to the higher noise and less details in infrared images, effects of general matching algorithms are not ideal. A novel infrared salient object stereo matching algorithm has been proposed. Salient object is extracted in both infrared images. And a sparse disparity map is generated by searching the homologous regions. Results demonstrate that the disparity of salient object is accurate and the speed is fast.Finally, an infrared salient object distance measuring system has been built. Passive ranging has been realized by utilizing the calibration results and disparity of salient object. Our distance measuring system has been experimented in real-world scenes and the error rate is below 15% for the distance of 10 meters, verifying the practicability of our system.