| Stereo vision and laser scanning are widely used in the field of UAV navigation. Consequently, inthis paper, some issues of stereo vision and laser scanning in UAV navigaiton are researched and thena scheme to measure scene3D depth based on the two technologies is designed.First of all, binocular stereo vision system is constructed. Internal and external parameters of thesystem are calibrated. The real scene images obtained from the system are rectified based on epipolarrectification.Secondly, considering the existing stereo matching algorithms it is difficult to obtain highmatching accuracy from images with radiometric differences, so to overcome this problem a novelalgorithm based on Census Transform and modified adaptive window is proposed. Experimentsdemonstrate that, comparing with the state-of-art local algorithms the proposed algorithm producesfairly matching accuracy and especially can handle the problem of radiometric differences which arenot solved by the state-of-art algorithms.Thirdly, for Census Transform, the intensity of pixels in the images is abandoned and therelationship between pixels is only reserved, which results in mismatch, a novel local method basedon joint matching cost strategy which is combined by Census Transfrom and color information ispresented. Experimental results demonstrate that the proposed algorithm can solve the problem ofmismatch and the error of the disparity result from images with radiometric differences is about only40%, comparing the algorithm above.Finally, analyzing the shortcomings of stero vision and laser scanning, we propose a scheme tomeasure scene depth based on stereo vision and laser scanning. By using region growing and EfficientGraph-Based Image Segmentation, the location of target can be determined precisely. Then, only thelocation needs to be scanned by laser scanner to obtain3D depth information. The scheme can makefull use of the two technologies and provide accurate target information for UAV navigation. |
