Study On Omnidirectional Camera Obstacle Avoidance And Measuring Of Application In Environment

Visual sensors own many advantages in the acquisition of information on the surrounding environment in field of mobile robot obstacle avoidance and navigation, such as abundant image information, multiple sensors' mutual non-interference.Consequently,visual sensors are used increasingly maturely and widely. However, the traditional vision has limited its application to a certain extent for limited observation range and incomprehensive environmental information. Omnidirectional vision has broad field and can make up the defects of traditional in the field of vision. Therefore, omnidirectional vision is widely using in autonomous robot navigation, three-dimensional reconstruction,video monitoring and other fields.Domestic and foreign experts and scholars have done a lot of research on the application of omnidirectional vision, but there are many deficiencies and defects:1. At present, majority of the platform used in the robot obstacles avoiding is single view omnidirectional camera and the obtained image is seriously distorted. Multiomnidirectional camera can get 360 degrees of images at real-time, and the distortion is small. Therefore, it is necessary to put forward a kind of stable, efficient and convenient ranging method for multi-omnidirectional camera.At the same time, our laboratory has achieved a certain performance in Bug obstacle avoidance algorithm. Bug obstacle avoidance algorithm is a kind of robot obstacle avoidance algorithm and requires a sensor with 360 degree detection range. Our laboratory has achieved a non 360 degree detection range of Bug obstacle avoidance algorithm, and can smooth the path to avoid obstacles to reach the end. However, the robot needs to be more frequent in the process of avoiding obstacles for the environmental information can not be obtained in all directions, which leads to the low efficiency of the obstacle avoidance.Therefore, this article is desire to replace laser range finder to sensor of 360 degrees and proposed a distance measurement which combines single vision and binocular vision based on omnidirectional camera Ladybug3.We study in detail how to realize the obstacle avoidance range by omnidirectional camera. In order to realize the ranging algorithm, this paper has done the following work:(1) The basic ranging knowledge of omnidirectional camera is studied, including camera calibration. After comparing several calibration methods, Zhangs' calibration method is chosen. Image preprocessing: By using histogram equalization and median filtering to enhance the image contrast and remove noise. Stereo matching: Several matching methods are analyzed. We decided to use improved SURF matching method to extract the feature points of coordinates for improving the matching of robustness.(2) The omnidirectional ranging principle of single vision combine with binocular vision is described. Firstly, we discussed how to determine the position of obstacle whether is in the overlapping area or non overlapping area. Secondly, the ranging mechanism which use binocular ranging in the overlapping region and monocular ranging in non-overlapping regions is determined. Thirdly,we realize the binocular distance measurement by perspective and the ranging principle is derived in detail. Fourthly, nonlinear regression modeling method was adopted to realize single vision ranging, and elaborates the specific establishing process of nonlinear regression model.(3) Completed the single and binocular vision experiments. Choose school playground as test site. We carry out the calibration experiment, the range image preprocessing, stereo matching experiment, binocular vision range test and the single vision range test. The experimental results and the error are compared and analyzed respectively.Experiments verify that the ranging method, binocular ranging error is in the range of1.08% to 4.48%, and the maximum error 4.48% appears in 4m. The error range of single ranging is 0.27% to12.57%, the maximum error 12.57% appears in 1.4 m. The change of the error is random, not as the distance increases. But it is significantly larger than the binocular ranging error. The above errors are in the acceptable range, and the results can be used to achieve obstacle avoidance.