Research On Obstacle Detection System Based On Binocular Vision

With the development of intelligent unmanned systems such as drones and robots,the requirements for obstacle detection and obstacle avoidance technology are also increasing.The use of computer vision for obstacle recognition has the advantages of high real-time performance,low cost,and low vulnerability to outside interference.Binocular stereo vision simulates human eyes,using two cameras to image the same scene from different positions to obtain images of the scene at different positions,and using the positional deviation of corresponding points between the image pairs to calculate the three-dimensional geometry of the target scene information.In order to achieve more accurate and fast obstacle detection and recognition in daily application scenarios,based on the principle of binocular stereo vision,this paper designed a detection system for deep recovery and obstacle segmentation of the target scene,which is the subsequent obstacle avoidance of the motion carrier.The work provides a technical basis.Through the analysis of the principle and mode of binocular vision and obstacle detection,this article selected the parallel binocular stereo vision mode as the overall design scheme;after considering the camera working environment,lens focal length,pixels and resolution and other factors,HNY was selected-CV-001 binocular camera was used to build the binocular stereo vision experimental platform in this design;afterwards,using Zhang Zhengyou calibration method,the binocular camera was calibrated using the calibration toolbox in Matlab and the distortion correction was performed by inputting the distortion coefficient to obtain The internal and external parameters of the camera;use the binocular camera to obtain the image and preprocess the image: first select the median filter and Gaussian filter to denoise the image,then use the histogram equalization method to eliminate the brightness difference of the image pair,and finally Use Laplace sharpening to process the image to enhance the details of the image;use the internal and external parameters obtained after calibration to perform parallel correction and image cropping on the binocular system,and analyze the differences between the three-dimensional matching algorithms through comparison and analysis,and then select the SGBM algorithm Perform stereo matching to obtain the target scene depth map.Finally,KMeans clustering algorithm is used for image segmentation to achieve obstacle recognition in the target scene.The obstacle detection system based on binocular vision designed in this paper is used to perform depth information acquisition and obstacle segmentation experiments on the target scene in the laboratory scene.Experimental results show that the system can realize the extraction and segmentation of obstacle depth information in multiple scenarios,and achieves the design goal.