CN-119625055-B - Shot three-dimensional space trajectory analysis and achievement prediction method based on 2D image

Abstract

The invention discloses a shot three-dimensional space trajectory analysis and achievement prediction method based on a 2D image, and relates to the technical field of computer vision. In order to fit multi-angle conditions, a 3D space curve model is built, a projection cost function is provided to calculate the difference value between the 3D space curve and points on a two-dimensional image plane, an optimization algorithm is used for correcting the 3D space curve to enable the 2D projection of the 3D space curve to coincide with the positions of a plurality of frames of shot, in addition, the position of a falling point is adaptively calculated in a mode of fusing a first falling bounce curve and a rolling track, and the relation between a physical distance and a pixel is automatically calculated according to an gravitation formula, so that the physical distance of actual falling is automatically calculated. According to the shot three-dimensional space trajectory analysis and achievement prediction method based on the 2D image, space curve fitting and high-precision landing point distance calculation are realized based on the 2D image, and shot tracking cost is greatly reduced.

Inventors

• LU YUXI
• ZHANG ZHUMING

Assignees

• 恒鸿达(福建)体育科技有限公司

Dates

Publication Date

20260512

Application Date

20241017

Claims (3)

1. 1. The shot three-dimensional space trajectory analysis and achievement prediction method based on the 2D image is characterized by comprising the following steps of: The method comprises the following steps of enabling a normal plane of a camera to be perpendicular to the ground, then obtaining a two-dimensional image of a process from the start of ball throwing to the stop of shot movement, detecting the position of the shot through a target detection model, and synthesizing a discrete track with a time stamp; Establishing a 3D space curve model, wherein the 3D space curve model is a cosine angle side casting model or a space oblique casting curve model, calculating a difference value between a 3D space curve and a point on a two-dimensional image plane through a projection cost function, and correcting the 3D space curve by using an optimization algorithm to the difference value so as to enable 2D projection of the 3D space curve to coincide with the positions of a plurality of frames of lead balls; The drop point prediction process comprises the steps of judging two inflection points in a discrete track so as to obtain an inclined-throwing curve track, a first rebound track and a rolling track; the actual distance calculating process automatically calculates the relationship between the actual height of the highest point and the pixel according to the gravity formula so as to automatically calculate the physical distance of the actual landing; the cosine angle side throwing model has the following formula: ; Wherein θ is the included angle between the plane of the shot flight track and the normal plane of the camera, and a, b and c are learnable parameters; The formula of the space oblique polishing curve model is as follows: ; wherein R is a space rotation matrix, As a learnable parameter, (x, y, z) represents the coordinates of the midpoint of the coordinate system; The formula of the projection cost function is: ; Wherein T 2d represents a point in the discrete track, the ith point is marked as [ x ti ,f(x ti )],P 3d represents a point on the fitted three-dimensional curve, the ith point is marked as [ x pi ,z pi ,f(x pi ,z pi ) ], and m represents the number of the discrete tracks; In the falling point prediction process, the falling point position is adaptively calculated according to a mode of merging a falling first rebound track and a rolling track, and the falling point prediction method specifically comprises the steps of fitting a 3D space curve model according to an oblique throwing curve track and the first rebound track, calculating an intersection point as a first intersection point position by combining two curves, obtaining a second intersection point position by combining a vertical regression line and the oblique throwing curve by using linear regression on the rolling track, calculating two norms of the first intersection point position and the second intersection point position, taking the first intersection point position as the falling point position when the Euclidean distance value of the two norms is smaller than a threshold value, and taking the second intersection point position as the falling point position otherwise; The accurate position p t of the highest point in the inclined throwing curve track and the accurate position p e of the falling point are calculated through a 3D space curve model, the time stamp t t corresponding to the p t and the time stamp t e corresponding to the p e are calculated through the relation among displacement, speed and time, the height value h of the highest point p t relative to the ground in the real environment is calculated through an gravitation formula, the linear mapping relation between the actual height and the pixel distance is obtained, and the actual distance D between the falling point and the falling point is calculated according to the linear mapping relation, wherein the formula is as follows: ; ; Wherein, the The pixel height that is the highest point of the curve, Is the pixel distance from the landing point to the starting point.
2. 2. The method of claim 1, wherein the optimization algorithm used for calculating the difference value of the projection cost function is a Levenberg-Marquardt optimization algorithm.
3. 3. The method of claim 1, wherein determining two inflection points in the discrete trajectory comprises calculating the inflection point using a discrete second order differential for three consecutive frames, the i-th point x i being the inflection point when: ; Wherein, the ; Where f (x i ) is the height of the ith point, Δ is the error threshold, Δx is the difference between x i and x i-1 .

Description

Shot three-dimensional space trajectory analysis and achievement prediction method based on 2D image Technical Field The invention relates to the technical field of computer vision, in particular to a shot three-dimensional space trajectory analysis and achievement prediction method based on a 2D image. Background In the context of AI sports development, the use of AI algorithms to achieve sports score calculation and comprehensive performance assessment is a trend. For shot movement, the traditional manual measurement distance and achievement analysis means are difficult to objectively, accurately and efficiently realize. Under the background, a low-cost and easy-to-lay scheme is needed to realize efficient and high-precision shot ranging. The existing method for realizing 2D image shot detection and track prediction based on the computer vision algorithm processing digital image is used for realizing prediction of the landing place at the image position by means of target detection, target tracking, quadratic curve fitting and other schemes, and is used for realizing physical displacement calculation of shot landing by means of placing markers for the image processing algorithm to estimate the relation between pixels and the actual physical distance after the distances are actually measured at a plurality of positions on the image in advance by means of a pre-calibration scheme. The scheme relying on 3D positioning needs to use three active beacons to perform spatial positioning based on RSSI, TDOA and other algorithms or spatial positioning based on image processing algorithms, so that accurate spatial position tracking and landing prediction are realized. The existing 2D image-dependent method cannot model a three-dimensional space, when the plane of a ball flight track thrown by an athlete is not perpendicular to the normal vector of a camera, the motion track shot by the camera is a non-standard quadratic function curve on a two-dimensional image, and although sampling points can be effectively fitted when a scheme such as a higher-order function or a complex multi-layer perceptron Model (MLPs) is used for fitting, under-fitting is easily caused under the condition of insufficient detection points of a complex scene, so that prediction errors occur. The method based on high-precision space positioning is usually realized by a triangular positioning algorithm such as Received Signal Strength (RSSI) or time difference of arrival (TDOA), and the method has high precision, but has higher requirement on the coordinate configuration of the beacon, and has higher and excessively complex practical deployment cost. Disclosure of Invention The invention aims to solve the technical problem of providing a shot three-dimensional space track analysis and achievement prediction method based on a 2D image, which realizes space curve fitting and high-precision landing point distance calculation based on the 2D image and greatly reduces shot tracking cost. The invention provides a shot three-dimensional space trajectory analysis and achievement prediction method based on a 2D image, which comprises the following steps: The method comprises the following steps of enabling a normal plane of a camera to be perpendicular to the ground, then obtaining a two-dimensional image of a process from the start of ball throwing to the stop of shot movement, detecting the position of the shot through a target detection model, and synthesizing a discrete track with a time stamp; Establishing a 3D space curve model, wherein the 3D space curve model is a cosine angle side casting model or a space oblique casting curve model, calculating a difference value between a 3D space curve and a point on a two-dimensional image plane through a projection cost function, and correcting the 3D space curve by using an optimization algorithm to the difference value so as to enable 2D projection of the 3D space curve to coincide with the positions of a plurality of frames of lead balls; The drop point prediction process comprises the steps of judging two inflection points in a discrete track so as to obtain an inclined-throwing curve track, a first rebound track and a rolling track; And the actual distance calculating process is used for automatically calculating the relation between the actual height of the highest point and the pixel according to an gravitation formula so as to automatically calculate the physical distance of the actual landing. The technical scheme provided by the embodiment of the invention has at least the following technical effects: The method comprises the steps of establishing a 3D space curve model, providing a projection cost function to calculate the difference value between a space curve and points on a two-dimensional image plane, enabling 2D projection of the 3D space curve to coincide with the positions of multiple frames of shot, solving the problem that the existing method relying on 2D images can not model a three-di