A Portable Golf Ball Tracking System Based On Stereo Vision Method

With the golf sport and the screen sports becoming more popular,portable golf sensors,which measures the the flying speed,launch angles and spin of the golf ball,are getting more attentions.Portable golf sensor has the advantages of small footprint and easy setup.It needs only one-time calibration.However,portable golf sensor faces more challenges than indoor sensor because it will be set up in varied scenarios in which the ambient light and backgound are complicated.In this paper,a portable golf sensor system based on stereo vision method is proposed to detect the position and capture the trajatory of the golf ball.Ball detection algorithms based on EDCircles and RANSAC are proposed to extract the image coordinates of the golf ball.Stereo calibration and 3d reconstruction are adopted to calculate the 3d world coordinates of the golf ball,which will be filtered by RANSAC line fitting method to remove false detected 3d world coordinates.Ball spin is estimated by adaptive threshold surface method.The proposed system has the advantages of high tracking accuracy,good robustness and low power consumption.Experimental results show that the system can track golf balls with speeds up to 90m/s and spins up to 11000 RPM with accuracy reaching the level of professional use.The main contributions of the paper include:1)Golf ball detection algorithmThe portable golf sensor has a variety of usage scenarios.When used indoors,the light intensity is around 300 Lux.However,the outdoor light intensity may be as high as 100000 Lux.The gray level difference between the ball and the background is relatively small,single-threshold binary algorithm is difficult to extract the ball position.This paper proposes an improved EDCircles method,combined with the contour filtering method and RANSAC method for golf ball detection.2)Golf ball track tracking algorithmThe portable golf sensor may capture many unwanted objects in the background,such as white shoes and a sky with a high gray value.When analyzing the trajectory of the golf ball,the sensor is susceptible to interference.In this paper,the initial ROI,hitting detection ROI,dynamic ROI are used to eliminate interference.In addition,this article designs a straight-line fitting method based on RANSAC to filter the 3D coordinates,so that the system is not affected by background interference.3)Calibration apparatusThe proposed portable golf sensor is based on stereo vision method,which requires calibration of internal and external parameters to calculate the ball coordinates in 3D world.This paper has designed a simple and effective one-time calibration method,using a stereo calibration apparatus to calibrate the cameras.4)Golf ball mark and mark segmentation algorithmIn low-contrast images,the ball markers are not much different from other pixels,and the single-threshold or multi-interval thresholds such as Chan-Vese and fuzzy entropy have low extraction accuracy.To estimate the spin of the ball,this paper proposes to use a mark extraction algorithm based on adaptive threshold surface,so that the golf mark can be segmented accurately.5)Implementation of portable ground golf sensor systemThis paper designs and implements the software and hardware of the portable golf sensor.It is implemented on a heterogeneous So C platform based on zynq-7z020.It uses FPGA to realize image acquisition,image preprocessing,and uses ARM processor to implement complex algorithms.With power consumption optimization,the system can achieve 8 hours of battery life powered by 10000 m AH battery.