Research Of Visual Recognition Algorithm For Service Robot Based On PCL

As we know, the intelligent of robots is largely dependent on the perception of the environments. The robots.make a decision to determine the next action by sensing the external arid the internal environments, which is the intelligent performance of robots. The robot vision is mainly about the perception of the external environments, with the functions of visual navigation, object recognition, positioning and so on through processing the image data captured by the sensor. The visual sensors are mainly industrial cameras and3D sensors, which respectively capture2D images and2.5D point cloud image. We can develop the image algorithm based on the open source vision library OpenCV and open source library PCL (Point Cloud Library). With the development of3D sensors, many scholars interest in the robot vision based on a3D sensor at home and abroad, and make a significant contribution to the field of the robot vision.In this paper, we studied the robot vision recognition algorithm based on PCL, built the point cloud model data library for common household objects. And we proposed a real-time recognition algorithm based on the shape feature of objects, analyzed the main algorithm processes of the point cloud processing steps, such as point cloud segmentation, feature extraction, and completed a human-computer interface design, and ultimately integrated the real-time recognition algorithm into the robot operating system ROS, implemented node communication, and analyzed the experimental results. The main work and contributions of this paper include:1. Developed a software platform for the common household objects building the point cloud cluster model. First we studied the common segmentation method and proposed a novel segmentation method based on the color and normal thresholds, then analyzed the characteristic of common feature descriptors in PCL, and selected CVFH feature descriptor as the realtime algorithm operator. Developed a software platform and built a point cloud model database for the common household objects.2. Designed a realtime object recognition algorithm based on PCL for the service robot. First, namely analyzed the merits of the characteristics of each algorithm in the pre-processing stage involving filtering, segmentation and feature extraction and other steps, and selected the optimal filtering, segmentation step of the algorithms. Finally, perfected the real-time recognition algorithm and analyses the real-time characteristic of the algorithm after some experimental verification.3. Successful integration of the object recognition system into the robot operating system ROS. First, builded an experimental platform to complete the realtime recognition algorithm based on ROS and verified the realtime rate and recognition rate of the point cloud recognition algorithm. Finally, verified the status of the recognition node communication with other functional modules nodes.Finally, summarized the full paper, analyzed the current problems in the study, and proposed the further research suggestions and prospects.