THE METHOD OF STEREOVISION ANALYSIS AND THE DEVICE FOR STEREOVISION ANALYSIS

A stereovision analysis device is made of a sensor, processing unit and projector is characterized by the fact that the sensor (1) is made of a housing (1.1), inside of which there are at least two RGB cameras (1.2) with variable focal length optics, positioned on a horizontal plane relative to each other in parallel, at focal distances ranging from 210 mm to 240 mm, connected by cables to a processing unit (3) with a power supply (3.1), at least one infrared illuminator (4) with control system (4.1) and power supply, where the sensor (1) is placed below the projector (5) and the projector (5) with the sensor (2) is placed in front of the subject in the working field of the cameras (1.2), and the processing unit is connected to an external computer (6), where the operation of the processing unit (3), communication with other devices and the implementation of the method according to the invention are carried out by dedicated software. The method of stereovision analysis is distinguished by the fact that the stereovision analysis device is started, then the focal distances in the range of 210 mm to 240 mm are determined, then the subject is placed in the working field, then the image registration process begins, then dedicated software using the SGBM algorithm, focal distances and viewing angles of the cameras and the images obtained from the cameras, calculates the distance of the subject in the working field from the camera system and creates a raster disparity map with a size equal to the size of the input image, then dedicated software using the MediaPipe BlazePose weave network, based on the two-dimensional images obtained from the cameras and the deep learning algorithm identifies and determines the location of the selected points of the subject, then the location of the selected points of the subject is superimposed on the depth map, then dedicated software using COCO model notation identifies the points describing the body of the subject, which form a convex quadrilateral, then dedicated software from the raster disparity map identifies the points lying within the quadrilateral and calculates their average distance from the cameras as an average for the next six image frames, then in parallel, the dedicated software using the notation of the COCO model identifies points outside the body of the subject, then the dedicated software from the raster disparity map identifies points lying outside the perimeter of the quadrangle and calculates their average