Font Size: a A A

Research On Handheld Object Pose Estimation Method For Assembly Process

Posted on:2024-09-04Degree:MasterType:Thesis
Country:ChinaCandidate:S Q YangFull Text:PDF
GTID:2568307073462154Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Target pose estimation is a hot research topic in the field of computer vision and has a wide range of applications in the fields of assisted assembly,robotic arm sorting and grasping,and human-machine collaboration.In the industrial assembly environment,it is important to study the target pose estimation method for handheld objects.In this thesis,a handheld object pose estimation method for the assembly process was carried out with the handheld object as the target.The method starts with the construction of a virtual assembly scene,and the handheld object pose estimation method for the assembly process was studied within the virtual assembly scene.The research focuses on simulation view data generation,view global feature extraction,feature index establishment,handheld object recognition detection and segmentation,and absolute and relative estimation of handheld objects.The work completed is as follows:(1)A library of standard template poses was created for the tools used in assembly operations.In order to solve the problems of difficulty and inefficiency in obtaining the actual parameters when generating the attitude library,a virtual assembly scene is generated using virtual reality rendering technology.The standard attitude view library is generated accurately and quickly by rotating the model with a resolution of 30° within the scene,extracting mean hash(aHash)features from the standard view,and building a feature index library based on a K-D tree.Used for initial pose search,improving search efficiency.(2)A handheld object segmentation method combining skin colour and hand shape features to extract hand regions,fused edge detection and simple linear iterative clustering(SLIC)presegmentation was investigated to address the problem of background interference in the assembly environment.The GrabCut segmentation method is improved by selecting the YCbCr skin colour model and combining the maximum inter-class variance method(Otsu)to identify hand regions,and combining hand biometric features to detect the position of fingers and palms as the target regions for segmentation.The handheld object is pre-processed using the Canny operator and SLIC super pixels,and then segmented using the GrabCut method.The segmentation result is then secondarily segmented and combined with the wavelet transform for image fusion to achieve segmentation of the handheld object.Compared with the original GrabCut algorithm,the segmentation efficiency is increased and the segmentation results are improved.(3)The absolute PnP pose estimation method with fused image features and the relative estimation method based on adjacent frames are investigated.The absolute estimation consists of two levels of coarse and fine estimation.The coarse alignment stage is based on the standard pose library search to obtain multiple poses,combined with principal component analysis(PCA)and cosine distance screening to obtain the initial pose;then the ORB algorithm is used to extract feature points and combined with GMS matching screening,and the PnP algorithm is used to optimize the coarse matching pose to complete the fine matching.The relative estimation stage solves the pose transformation of adjacent frames based on the pairwise geometric constraints.For the problems that the absolute estimation process is timeconsuming,and the relative estimation will accumulate errors,a joint absolute and relative estimation pose estimation method is designed.The result of the absolute estimation is used as the first frame of the relative estimation,and the target pose is solved continuously based on the relationship of adjacent frames.By setting the number of frames of successive estimation to clear the accumulated error at regular intervals,the joint estimation improves the real-time performance of the system compared with the absolute estimation.Finally,this thesis implements the pose estimation method for handheld tools through the constructed virtual assembly scenario,and tests the pose estimation method in the virtual scenario with the handheld object as the target.The experimental results show that the handheld object pose estimation method for the assembly process in this thesis can correctly identify the pose of the handheld object,and improve the real-time performance of the method by combining absolute and relative estimation.
Keywords/Search Tags:Pose Estimation, Virtual Assembly, GrabCut Segmentation, Handheld Object, PnP Algorithm
PDF Full Text Request
Related items