Research on Image Lightweight Binocular Ranging

Aiming at the problems of low speed and accuracy of traditional binocular ranging methods, this paper discusses the influence of a filtering sampling preprocessed lightweight image of binocular ranging. Firstly, the image to be detected is processed with different levels of Gaussian filtering and up-down sampling, and different levels of preprocessed images are obtained. Secondly, based on YOLOv5 object detection and recognition, MATLAB binocular calibration and stereo rectification are carried out, and then SGBM and BM stereo matching algorithms are respectively combined to realize the binocular ranging of the two methods. Finally, the relationship between binocular ranging range, speed, and accuracy is analyzed under the condition of image preprocessing. The results show that the image preprocessed by one upsampling and downsampling can achieve a simultaneous improvement in the speed and accuracy of the range measurement.