检测斯诺克台球 [英] Detect Snooker Billiard Balls

问题描述

我正在使用OpenCV启动一个项目，以准确检测斯诺克球(台球)的位置和轮廓.

I am starting a project with OpenCV to accurately detect the position and contours of snooker balls (billiard balls).

这些是球:

我尝试了几种算法，但没有用.到目前为止，这是我所做的:

I've experimented with a few algorithms to no avail. This is what I've have done so far:

- HoughCircles :这几乎行不通.即使将参数调整为毫米分辨率，也只能检测到几个球，并且不能准确地检测到球的中心.

-HoughCircles: This barely works. Even adjusting parameters to millimeter resolution only results in a few balls detected, and the center of the balls are not accurately detected.

- Canny :在这里，我只能在球彼此不靠近的情况下检索球的轮廓.

-Canny: Here I only can retrieve ball contours if the balls are not close to one another.

- HSV颜色阈值:因为球不是一种颜色，所以这是行不通的.

-Threshold by color HSV: Because the balls are not one color this just doesn't work.

我想尝试其他算法，例如HaarCascadeDetection或SURF.

I want to experiment with other algorithms like HaarCascadeDetection or SURF.

你们怎么想?

推荐答案

仅使用霍夫圆变换(在使用参数一段时间后)，我得到了一些不错的结果.我猜想诀窍是结合使用足够小的param2(以便进行更多检测)和限制搜索范围(以滤除误报).

I got some good results using just Hough circle transform (after playing with the parameters for awhile). I guess the trick is the combination of using a small enough param2 (to allow more detection) and restricting search radius (to filter out false positives).

circles = cv2.HoughCircles(img,cv2.cv.CV_HOUGH_GRADIENT,1,5, 
            param1=100,param2=10,minRadius=6,maxRadius=10)

我也消除了阴影，但是我不确定在您的情况下是否需要它. 这是一些测试结果.

Also I removed the shadows, but I'm not sure if it's needed in your case. Here's some test results.

这篇关于检测斯诺克台球的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！