Scene segmentation and target modeling method based on concavo-convex and RSD features

Abstract

The invention discloses a scene segmentation and target extraction method based on concavo-convex and RSD features, and belongs to the technical field of computer vision. The method comprises the following steps: step 1, acquiring a scene point cloud by using Realsense and a PCL, and performing radius filtering; step 2, over-segmenting the scene point cloud based on hyper-body clustering; step 3,completing scene segmentation based on concave-convex re-clustering by using an LCCP method on the basis of the superbody clustering obtained in the step 2; and modeling the target object according tothe RSD geometrical characteristics of the segmented scene point cloud target object. Experimental results show that the target object modeling problem in any scene can be effectively solved. The method is a learning-free method, and has the advantages of low calculation time consumption and high occasion applicability.

Description

technical field [0001] The invention belongs to the technical field of computer vision, relates to the grasping suitable for robots in any scene, and can be used in the fields of unmanned driving, unmanned aerial vehicles, deep space detection, underwater robots and the like. Background technique [0002] With the development of science and technology, robots are becoming more and more intelligent. Currently, robots pursue autonomy and high intelligence. When a robot performs autonomous or intelligent tasks in an unknown environment, it needs to understand the environment. It needs to recognize and understand objects and objects in the current field of view or environment like the human brain, and perceive its own environment. Image segmentation technology is the key technology to enable robots to perceive and recognize the environment. With the rapid development of laser 3D ranging technology, there have been many 3D cameras that can collect 3D information in 3D space scen...

AI Technical Summary

Problems solved by technology

Another class of algorithms uses learning-based methods combined with optimization of conditional random fields and Markov random fields to deal with segmentation problems, but their high computation runs limit their use in practical applications

The shortcomings of the above methods have greatly affected the subsequent target object modeling and the robot's grasp of the target object.

Images