Research Of Binocular Visual Positioning On Tele-operated Construction Robot Based On Neural Network

Tele-operation construction robot is the robot system used to complete the complicated work in the dangerous environment by remote operation. In order to improve the working efficiency of tele-operation construction robot, tele-presence technology tips is used on the engineering, like visual, auditory, haptic, motion perception. Visual cues technology is to capture images on live by cameras and transmit those images to the operator through the network, which the operator complete the complex operation by image information. However, because of the large amount of image data, there is a big delay, which may cause the construction robot can’t work normally. In order to solve this problem, it is necessary to obtain the positioning information of the robot by taking three cameras stereo visual positioning technology. But the three camera stereo visual positioning is complex, high cost compare with binocular stereo visual system. Therefore, the binocular stereo visual technology for tele-operation construction robot is studied.First of all, according to the requirements of the construction robot vision positioning accuracy, I built a tele-operation construction robot visual positioning hardware platform by choosing camera and lens, designed a tele-operation construction robot binocular vision positioning software platform, wrote the man-machine interface based on MFC, and captured images by using thread.Secondly, I determined the relationship between the four coordinate conversion camera, analyzed the causes of image distortion, calibrated camera based on Halcon, got the camera internal parameters, and corrected the image distortion.Then, I analyzed methods used to extract feature commonly, studied the detection algorithm about feature extraction. I set features on the tele-operation engineering robot, extracted feature by using morphological method and template matching method. Compared these two kinds methods of feature extraction, the experiment results show that the stability of extract characteristic data based on template matching is slightly better, and finally selected the template matching method.Finally, I studied the BP neural network algorithm, established network model on tele-operation construction robot visual positioning system, optimized the weights and threshold of BP network model by genetic algorithm, realize visual positioning of tele-operation construction robot by using network optimization model.The experiments show that the method of binocular stereo vision based on neural network is simple and effective, the positioning accuracy can meet the requirements of construction robot, visual positioning of tele-operation construction robot based on the neural network optimized by genetic algorithm improved the precision and had good robustness, which can provide technical reference for tele-operation construction robot position feedback under the environment of network.