Remote Measurement System Based On Binocular Stereo Vision

With the continuous development of computer vision theory,binocular stereo vision measurement technology is widely used in vehicle distance measurement,robot obstacle avoidance,virtual reality and military fields.The technique uses two cameras with the same parameters to simulate the visual mechanism of the human eye.Then the system can acquire two images in different perspectives in the same scene,and use the principle of parallax to establish a mathematical model to recover the depth of three-dimensional objects.In the existing research of the binocular stereo vision system,most systems mainly focus on the identification of obstacles,and there are few studies on the accuracy of ranging.Most system parameters are selected based on experience,and there are few theories that can be used for reference.In the actual measurement process,the system needs to select system parameters according to the limitations of the distance to be measured and the actual application scenario.At the same time,No matter how far the obstacle is,the accuracy of the distance measurement will be affected by the installation errors of the left and right optical axes in the system.In engineering applications,all kinds of errors including installation errors,system errors,parameter errors need to be corrected to improve the accuracy of distance measurement.This paper aims to improve the precision of long-distance ranging system.Firstly,the basic principle of binocular stereo vision is studied,and the existing binocular stereo distance measurement model and its characteristics are analyzed.Then,the main research is to improve the accuracy of long-distance ranging systems.In view of the common parallel distance measurement model,the error of ranging measurement caused by the non-parallel optical axes of the right and left cameras in engineering applications was studied.Using the principle of geometrical optics,the influence of structural parameters to ranging measurement such as focal length,baseline and object distance were analyzed.And the relationship between the structural parameters and ranging error of the ranging model is quantified to guide the choice of system structure parameters.In view of the problem that the optical axes of the left and right cameras are not parallel in the distance measurement process,Bouguet image correction algorithm based on the epipolar geometry constraints is used to correct the left and right images to recover the ideal parallel distance measurement model.In view of the problem that the calibration accuracy of the camera calibration algorithm is not high in the outdoor long-distance application and the model cannot be established accurately,Zhang Zhengyou's calibration algorithm is selected in this paper and the theoretical analysis and multiple outdoor experiments are conducted to summarize the constraint conditions of long-distance calibrations.Finally,the calibration accuracy of longdistance ranging systems can be achieved.Due to factors such as the manufacturing process and installation errors,it is difficult to keep the two camera parameters consistent exactly,resulting in a decrease of the accuracy of measurement in the actual system.In this paper,the original ranging formula has been revised to adapt to different camera parameters to ensure accuracy of long-distance ranging and reduce the error of measurement.Based on the previous research work,an experimental platform for long-distance ranging measurement was set up.Then,the code of related software algorithms was coded,and experiments were conducted to verify the effectiveness of the method.The experimental results show that when the distance is less than 60 m,the measurement error is less than 3% and when the distance is from 60 m to 100 m,the distance measurement error is less than 8%,which means the accuracy of long-distance ranging systems can be improved effectively.