A Trajectory Planning Method for Robot Scanning System Using Mask R-CNN for Scanning Objects with Unknown Model

Laser scanning has played an important role in many fields of industrial production, such as product defect detection, reverse engineering and so on. Automatic laser scanning is the key technology to improve scanning efficiency and accuracy. This paper presents a method for planning laser scanning trajectory, especially for the objects with unknown model. This method is also suitable for painting and welding operations. Mask R-CNN is used to segment RGB images. Combined with the well-registered depth map, rough point cloud of processed objects can be extracted, which can be used as the basis of trajectory planning. In the part of trajectory planning, after extracting the outline of objects from rough point cloud information, the least square method is used to smooth the trajectory. In the actual scanning process, a method of real-time adjusting sensor position and posture by using PID controller is proposed to optimize the collected data. Finally, we construct an actual scanning system to verify the effectiveness of our method.