Research On Target Recognition And Location Of Robot Based On Monocular Vision

The vision has some advantages of abundant information, high precision and wide detection range, especially the advantages in cost, calibration and robustness. It is widely applied in many fields such as objection recognition, industrial inspection, navigator of roboties. Therefore, for the monocular industrial robot, the paper studies that the industrial robot can complete target recognition and location in many workpieces coexisting situation to realize intelligent crawling task, especially real-time match recognition technology and space position technology in detail. The main task is as follows:Firstly, in order to improve the image quality of the workpiece uptaked from the CCD camera, the original image should be preprocessed, including median filtering and segmentation. For the workpiece image, various pretreatment methods are analysed by comparing the experimental results. Then, for color image, an improved segmentation method- watershed segmentation based on grid is proposed. Experimental results show that the method can effectively solve the over-segmentation and remove isolated points.Secondly, in view of the workpiece recognition, an algorithm based on wavelet moment is proposed. Wavelet moment combines wavelet analysis with moment invariants, thus it not only has the character of scaling, rotation and translation invariance, but also improves the master degree of fine structure of the image and reduces the noise influence. Experimental results show that the algorithm is easy to be achieved, and it can efficiently detect the target in the motion of parallel displacement, rotation or microscale change. The recognition accuracy and anti-noise capability of wavelet moment have been much increasing.Thirdly, for target location, the article mainly describes target positioning principle in the single camera and relevant calibration methods. Target centroid is extracted for locating. Considering the particularity in industry and no distorion, Zhang Zhengyou plane template two steps is used and the method makes the calibration experiment much easier with high precision. Finally the locating parameters are obtained, and the feasibility of the method is demonstrated.Last, by using the recognition and position methods referred above and the hardware system, which consist of GRB-400 robot, CCD camera, image sampling card and PC, grabbing component experiment is accomplished.