Human Hand Feature Recognition And 3D Reconstruction In Human-machine Assembly Environment

Automated information processing algorithms that can adapt to multiple scenarios are effective and practical means to improve work efficiency,manufacturing system flexibility and product quality.The human-computer interaction cooperative assembly technology based on binocular machine vision uses the hand position posture obtained by visual processing as the input information of the assembly robot behavior decision.Achieve efficient and flexible production processes through the cooperation of assembly tasks.The biometric information of the hand image and the information related to the hand gesture associated with it are the premise of the hand position prediction.This research has high research value and practical value.In this paper,through the theoretical analysis,extraction and recognition of hand biometrics in human-computer interaction and assembly images,binocular 3D reconstruction and inference,the hand part posture reconstruction based on hand features is realized.On the basis of skin color modeling,the image skin segmentation is used to realize the image localization of the hand position,and the inner contour line of the hand is recognized,which provides a basis for the biometric recognition of the hand.In the YCrCb color space,the maximum inter-class variance method is used for adaptive threshold recognition to achieve basic positioning and general contour acquisition of hand images.A skin color mixed Gaussian model is established on the RGB color space based on the clustering learning method.The accurate segmentation of the hand region is realized by the expectation maximization algorithm with the adaptive threshold recognition result of YCrCb color space.A more accurate hand profile is obtained.The texture points are extracted on the human hand grayscale image based on the improved Laplace operator,and the contour of the hand shape is obtained.Combined with the hand-shaped outer contour as a complete hand contour,the complete contour of the human hand is recognized.Identify and infer the biometric features of the fingers,extract the complete phalanx,identify the location of the proximal and middle phalanx,and infer the location of the distal phalanx and the position of the wrist.The feature detection of hand structure is realized.The outline of the hand shape and the inner contour are represented by the image data as a contour coordinate chain in a sequence search manner;The finger root position is identified by the contour of the hand shape based on the regional corner algorithm Combining the contour of the hand and the contour of the hand into a complete contour of the hand,and identifying the fingertip position by the complete contour of the hand;The finger root and fingertip recognition results are corrected by the recognition result of the inner contour of the hand shape.The finger root and the fingertip recognition result are used to extract the substitute finger root point,and the finger shape is extracted by the finger root and the fingertip point;The position of the proximal and middle knuckles is extracted based on the improved Laplace operator.Inferring wrist points using the alternate finger root point in combination with the hand contour.The far knuckle position is inferred on the image with the extracted finger shape and the near and middle phalanx positions.Integrate all the information identified and inferred to achieve hand feature detection.Three-dimensional reconstruction and pose inference of the biometric features of the opponent's parts,and the complete three-dimensional spatial pose information of the hand is obtained.The three-dimensional pose of the biometric features of the hand was obtained by linear reconstruction of the biometric features of the hand.The hand position model is established based on the length of the phalanx,and an adaptive threshold inference algorithm for spatially inferring the distal knuckles is established.The part of the three-dimensional reconstruction information is brought into the human hand pose model,and the three-dimensional information of the far knuckle is inferred from the space.Integrating the three-dimensional information of all hand biometrics,the hand position information in the three-dimensional space is obtained.