Design And Implementation Of The Cleaning Robot’s Obstacle Detection Based On Binocular Vision

Cleaning robots attract more and more attention with the popularity of intelligent household concept in recent years. Domestic research in this field is still in the stage of development, which has much room for improvement. The obstacle detection system plays an important role in the cleaning task. So the research has important theoretical and realistic significance. Traditional obstacle detection methods like infrared and ultrasound have their inherent drawbacks, such as the narrow signal detection range, incomplete detection information and the limits of system expandability. This paper plans to introduce vision sensors to cleaning robot’s obstacle detection system, and designs the obstacle detection algorithm based on the theory of binocular stereo vision.This paper describes the research status of cleaning robot firstly, and analyses the implementation principle of their detection systems. In addition, this paper introduces the development status of visual obstacle detection methods and specifies the main research content. Secondly, the system’s functional requirements are proposed based on the survey of literature and market demands. Finally, the overall scheme of cleaning robot’s obstacle detection is designed.Stereo calibration is the basis of obstacle detection method based on binocular stereo vision. Through the camera imaging model, the transformation relationship between the image coordinates and the 3d world coordinates is built. The distortion of camera lens also needs to be calibrated. In this paper, the intrinsic and extrinsic parameters of left and right camera are obtained on the basis of Zhang’s calibration method. Then, we complete stereo calibration by binocular camera model.Real-time obstacle detection is crucial for the normal work of cleaning robots, and stereo matching is the most complex part of the binocular visual obstacle detection scheme. Therefore, the improvement of speed in stereo matching is the key point of this paper. This article selects region matching algorithm as the basis of research, and improves the efficiency of matching by combining multi-resolution matching and adaptive search scope by reducing the redundancy calculation. Through the test of disparity validity to reduce wrong matching, a new fast region matching algorithm is proposed finally, which can improve the matching speed within a certain error range.The depth information in disparity map is used to extract obstacle area. In this method, Barriers on the surface and outside the route are removed at first. And then, the objects in the foreground are extracted by threshold segmentation. The target obstacles can be obtained through the area information of connected regions. In addition, the paper tries to transplant the algorithm to the embedded platform of ARM+Linux, which promotes the practical process of binocular stereo visual obstacle detection method.Each function module is verified through experiment with detailed results. Ultimately concluded, the proposed obstacle detection method in this paper can detect obstacles on the path quickly and reliably.