Research On Multi - Vision Fusion Model Of Industrial Robot

Visual perception is an important embodiment of intelligent industrial robots, and it is one of the core technologies of robotic application to flexible manufacturing. Since the mid-1990 s, our country has been carrying out research in the field of intelligent robots, and significant progress has been made in robotic feeding, machining, welding, assembly, inspection and so on, which strongly promotes the development of intelligent industrial robot technology. However, the vision research and application of the domestic industrial robots is concentrated on monocular or binocular active vision system. When working in complex environment where industrial robots need measurement, tracking, obstacle avoidance, target acquisition and positioning and other operations, it is difficult for industrial robots to complete reliably these tasks only relying on the monocular vision. By using of multi-camera vision perception, it can improve the reliability, speed and accuracy of robot operation, especially when the robotic active vision combining with passive vision forms a multi-dimensional visual feedback, it can help industrial robots complete effectively the complex visual tasks.Considering the limited field of view of conventional robotic hand-eye vision and absences of visual feedback robotic arms, a multi-ocular vision system is established for industrial robot by combining passive environmental vision based on side and top-view cameras with active vision on robotic arms, and in the system each camera is an agent with the ability of image processing and analysis, these agents form a multi-ocular vision of industrial robot based on multi-agent-system. By carrying out research on multi-ocular perception and motion tracking model about industrial robots, a multi-ocular vision fusion model suitable for industrial robots is presented based on the MAS structure, combined BP neural network and D-S evidence theory. The MAS structure is used to deal with multi-ocular images cooperatively. Combined BP neural network is used for visual fusion of multi-ocular vision images and multi-feature fusion of single image. D-S evidence theory is used for decision fusion of the output of the BP neural network.The research on recognition of typical parts and industrial robot manipulator tracking shows that: the reliability of target recognition and the accuracy of target location are improved by using multi-ocular vision. The cooperation among agents in MAS speeds up the image processing and recognition, which improves effectively the real-time performance of the robotic multi-ocular vision.