Research On Pose Estimation Of Robot And Target Object Based On Computer Vision

With the development of technology,the tasks undertaken by the robot are becoming more and more complex,which requires the robot to accurately perceive the information of environment,and the relative pose between the robot and the target object.It is a necessary condition for the robot to complete relevant tasks,such as sorting robot,assembly robot,etc,which need to be guided by the relative pose between the robot and the target object.The premise of pose estimation is to accurately detect the target object.Therefore,object detection and pose estimation have important research significance and application prospects in the field of robot.Computer vision is a common method to study object detection and pose estimation,but the existing image processing methods are not ideal for weak texture and non texture image processing.Aiming at the above research problems,this paper studies the problem of object detection and pose estimation in weak texture images by using the image features of regional structure tensor.The main research work is as follows:(1).In order to better process weak texture images,this paper proposes a feature point algorithm:Image Structure Information Features(ISIF)algorithm.The traditional feature point algorithm focuses too much on the local texture features of the image,and lacks the investigation and description of the surrounding area and global structure information.The ISIF feature descriptor uses the regional structure tensor of the image,which can reflect the regional structure information and local texture information of the image to a certain extent,It has a good detection effect for weak texture images.By constructing the scale pyramid,the ISIF descriptor has scale invariance.At the same time,the relative vector is used to describe the feature information of feature points,which makes ISIF have good rotation invariance.Then,ISIF algorithm test experiments are carried out,including ISIF feature descriptor performance test experiment,parameter sensitivity experiment and occlusion test experiment,which verify the effectiveness and robustness of ISIF algorithm.(2).A pose estimation method based on the information of three-point is proposed(PETP).This method makes full use of the relative position of feature points and objects,the corresponding relationship between feature points and the three-dimensional position information of feature points to calculate the pose of objects.Compared with the traditional point cloud registration pose estimation algorithm,the information is used more comprehensive,and the pose estimation results are more suitable for the requirements of robot operation.Then,PETP test experiments are carried out,including the feasibility experiment and parameter sensitivity experiment of PETP,which verifies the effectiveness of PETP.(3).The pose estimation algorithm of the robot based on PETP and the ISIF feature point algorithm is proposed.The image feature points are extracted and matched to realize the target object detection in the weak texture image.Based on the feature point matching information,the pose estimation of the target object is completed by using PETP.(4).Carry out the experiment of robot vision system,build a vision system for robot operation,design each module of the vision system based on the previous algorithm research,and finally test each module of the system to further verify the effectiveness of the algorithms proposed in this paper.Aiming at the problem of pose estimation of target object in weak texture image,this paper proposes ISIF feature point pose estimation algorithm,and theoretically deduces and studies the algorithm according to the construction idea of the algorithm.It is verified theoretically that ISIF feature point pose estimation algorithm can solve the problem of pose estimation of target object in weak texture image.Based on the algorithm principle,the local algorithm test experiment and system experiment are carried out.Based on a large number of local algorithm test experiments,the effectiveness and robustness of ISIF feature point algorithm,object detection algorithm and pose estimation algorithm are verified.Finally,the experiment of robot vision system is designed to verify the effectiveness of ISIF feature point pose estimation method.