Automatic Target Reporting System Based On 3D Machine Vision

With the rise of the national fitness boom,archery has attracted more and more attention and love from sports enthusiasts.In archery,target reporting is essential.The method of traditional manual target reporting has disadvantages such as long-time consumption and poor safety.Compared with manual target reporting,automatic target reporting can be a solution to these disadvantages,and has a great accuracy.The existing hardware-based automatic target reporting systems mainly utilize sensors to modify the target surface,which depend strongly on the accuracy of sensors.The automatic target reporting systems based on image processing mainly handle the captured video frames.Such systems do not need to modify the target surface,but they cannot effectively recognize the situation that the arrow hitting points are obscured and the arrow hitting points of new arrows are mixed with the arrow holes left on the target surface.To address the problem of the existing automatic target reporting system,this thesis proposes an automatic target reporting system based on 3D machine vision,combining binocular stereo vision and image processing technology.This thesis utilizes binocular vision to reconstruct 2D arrow target images in 3D,which can enable problems such as arrow occlusion to be effectively solved.The main work is as follows:1.A method of extracting key frames of archery video based on frame interval difference and arrow stability check is proposed.Inter-frame difference is used for the extracted adjacent frames,and then whether there are new arrows is determined in the V channel..When the presence of new arrows is detected,a delayed verification method is used to ensure that the new arrows in the key frames are in a stable landing state.2.A method of obtaining arrow landing points in 3D space based on binocular vision is proposed.The accelerated CREStereo network is used to obtain the parallax information of key frames,and then the point cloud data of the target surface and the arrow are generated according to the geometric principle of binocular imaging.Afterward,the point cloud of the arrow is fitted through the approach of least squares to obtain the complete arrow information.The set of common points of the target surface and the arrow is the landing position of the arrow.3.A method of Accurate identification and verification of loop numbers calibration based on the traditional image segmentation algorithm is proposed.The ring information is obtained through the algorithms of color segmentation and ellipse fitting,and then the arrow point data in 3D space is remapped into the 2D image to determine ring numbers according to the position relationship between the arrow point and the ring line.When the arrow point is close to the loop line,accurate verification is required.The RSF model is used to segment the close ring line and detect whether the arrow point intersects with the ring line to ensure the accuracy.4.A software system of automatic target reporting based on the method proposed above.It includes automatic target reporting module,3D display module,information management module and score statistics module.