A stereo vision depth estimation method of binocular wide-field infrared camera

With the development of binocular stereo vision ranging technology, traditional visible light cameras which have a narrow field of vision gradually cannot meet the need of some special environments. For this reason, this paper proposes to use large field of view infrared cameras to replace traditional visible light cameras. Aiming at the distortion defects caused by the large field of view lens and the low calibration accuracy of the infrared camera, an active infrared radiation calibration board is used. The internal and external parameters of the camera are accurately calibrated under Taylor model which is different from the traditional pinhole model. In addition, we use the SIFT+RANSAC algorithm for feature point extraction and matching, and then get the depth information. The experimental results of target ranging at different distances show that when the distance reaches 9m, the error rates of the Taylor model (7.11% and 10.1%) are better than the pinhole model (16.4% and 29.8%). The image correction effect of the Taylor model calibration is much better than that of the pinhole model, and the accuracy of depth estimation is also better than that of the pinhole model.