Research On Multi-source Visual Perception For Industrial Tri-Co Robot

Traditional industrial robots are characterized by high-speed,high-power and lack of perception ability,so they are often isolated from their surroundings while working.Industrial robots with the ability to perceive have lower requirements of their working atmosphere,therefore,they can work safely and efficiently with human workers.Industrial tri-co robots have wider application scenarios and higher work efficiency compared with traditional industrial robots,and are more competent for complex and changeable production and manufacturing tasks in the future.Rapid and accurate perception of the environment and staff is helpful for industrial tri-co robots to adopt different operation strategies according to the changes of environment and status of stuff.This paper makes a deep study on the multi-source visual perception of industrial tri-co robots to provide a vision perception method which offers abundant perception information.The main contents of this paper are as follows:(1)Research on multi-source visual data acquisition and pre-processing methods.Combining with the hardware structure and working principle of Kinect sensor,the transformation from the sensor's raw sensing data to three-dimensional point cloud sensing data containing human,object and color information is completed.Based on the analysis of the noise characteristics of Kinect sensor,a point cloud denoising method is proposed.The proposed method is of excellent performance in point cloud denoising and runs fast which guarantees the real-time capability of visual perception.(2)Research on real-time fusion processing method of multi-source visual data.A new method of sensor external parameter calibration based on iterative closest point(ICP)point cloud registration algorithm is proposed.A simple and easy-to-make calibrator is proposed to be used as the target of point cloud registration,which makes the algorithm result more accurate and credible.Compared with traditional methods,this calibration method not only ensures accuracy,but also has the advantages of convenience and easy operation.Compared with real-time ICP registration,fixed multi-view point clouds can be transformed and fused using the results of external parameter calibration,which has faster running speed.The perception range of fused multi-view visual data is wider and the anti-occlusion ability is stronger.On the basis of the fused point cloud,the perceptual information is refined,and the bounding boxes of people and objects are computed in real time based on principal component analysis.Human workers and obstacles are treated differently,so as to improve the efficiency of human-computer integration on the premise of ensuring the safety of collaborators.(3)Design and implementation of multi-source visual perception prototype system for industrial tri-co robot.In the tri-co manufacturing unit based on IRB 1200 industrial robot,the dynamic acquisition,transmission and fusion of the sensing data from Kinect sensors are realized by using multiple PCs through local area network.Combined with the real-time generated bounding box data and the acquisition of the working status of the robot from the robot controller,tri-co decision is made quickly,and human-robot cooperation under simple tasks is realized.