Research On Target Recognition And Tracking Technology Of Photographic Robot Based On Machine Vision

This subject is derived from a project commissioned by enterprises and institutions.It is responsible for the development of the vision module of the virtual studio photography robot.It mainly implements the system development of the recognition and tracking of the target host by the photography robot.Tracking module.This paper first analyzes the working space requirements of a photographic robot in a virtual studio,proposes an orbital photographic robot,and introduces its structure and control system in detail.This paper analyzes the kinematics of this camera robot and solves its inverse kinematics.In the virtual studio,all guests except the host are highly mobile.At this time,the face recognition problem that needs to be completed can be abstracted into the face recognition problem on the open set.Based on the realization of two classic face recognition algorithms,this paper implements them and analyzes their advantages and disadvantages,and then proposes a face recognition model based on YOLOv3 and improved Res Net50 network.The improved recognition network is jointly supervised and trained by the central loss function and the Softmax loss function.Finally,experiments have proved that the algorithm has better recognition accuracy and real-time performance on the open set,and can be applied to the face recognition module of photographic robots..The target tracking of a photographic robot cannot be separated from the spatial positioning of the tracking target.This paper first analyzes the camera imaging model of the photographic robot,and completes the internal calibration of the camera based on Zhang Zhengyou's chessboard calibration method.Based on the cascade regression tree,this paper completes the detection of the feature points of the face,and then proposes a depth positioning algorithm based on the pupil distance and similar triangles.Considering the impact of head posture on positioning accuracy,this paper has further improved the algorithm by introducing a scalar to measure the head posture change and correct the pupillary heart distance,so that it can still have stable robustness when the head posture changes Depth positioning accuracy.Based on the kinematics analysis and proposed face recognition module and depth positioning module of the camera robot,this paper proposes an ideal target recognition and tracking solution for the camera robot,and derives the three-dimensional coordinate formula of the tracking target..Based on the spatial coordinates of the tracking target and the inverse kinematics solution,the joint variables of the camera robot can be obtained in the ideal tracking state.Because there are some problems with the tracking scheme in the ideal state,based on the tracking requirements in the virtual studio,the target tracking bounding box is preset,and two camera robot tracking modes are provided according to the target offset distance,and verified in experiments.The tracking scheme has high tracking accuracy,good visual picture,and stable mechanical performance of the photographing robot during tracking.