如何实现正确率比较高的颜色识别

一只人

最近弄一个还原魔方的机器，但是颜色识别不太理想，主要是灯光影响太大，识别率并不高，也就60%-%70，波动比较大。我用的就是转换成hsv进行卡值，就是大概在哪个范围属于哪个颜色的这种处理方法。有没有更加智能一些的方法，求指教。能不能用一些机器学习的算法，来进行动态的学习。这一块不是很懂，如果这方面可行的话，打算去学一下机器学习。

smallyouyou

用類神經做看看，
這邊有基本的用法，
我還沒用過opencv這塊，
可能要研究一下，
只能先找了這個網址給你參考用法，
http://blog.csdn.net/xiaowei_cqu/article/details/9027617
然後下面是貝氏分類器的例子也可以做看看，

1. #include "opencv2/opencv.hpp"
   
2. #include "iostream"
   
3. 
   
4. 
   
5. using namespace cv;
   
6. using namespace std;
   
7. 
   
8. double inputArr[10][13] =
   
9. {
   
10.   1,0.708333,1,1,-0.320755,-0.105023,-1,1,-0.419847,-1,-0.225806,0,1,
    
11.   -1,0.583333,-1,0.333333,-0.603774,1,-1,1,0.358779,-1,-0.483871,0,-1,
    
12.   1,0.166667,1,-0.333333,-0.433962,-0.383562,-1,-1,0.0687023,-1,-0.903226,-1,-1,
    
13.   -1,0.458333,1,1,-0.358491,-0.374429,-1,-1,-0.480916,1,-0.935484,0,-0.333333,
    
14.   -1,0.875,-1,-0.333333,-0.509434,-0.347032,-1,1,-0.236641,1,-0.935484,-1,-0.333333,
    
15.   -1,0.5,1,1,-0.509434,-0.767123,-1,-1,0.0534351,-1,-0.870968,-1,-1,
    
16.   1,0.125,1,0.333333,-0.320755,-0.406393,1,1,0.0839695,1,-0.806452,0,-0.333333,
    
17.   1,0.25,1,1,-0.698113,-0.484018,-1,1,0.0839695,1,-0.612903,0,-0.333333,
    
18.   1,0.291667,1,1,-0.132075,-0.237443,-1,1,0.51145,-1,-0.612903,0,0.333333,
    
19.   1,0.416667,-1,1,0.0566038,0.283105,-1,1,0.267176,-1,0.290323,0,1
    
20. };
    
21. 
    
22. //一个测试样本的特征向量
    
23. double testArr[11][12]=
    
24. {
    
25.   0.125,1,0.333333,-0.320755,-0.406393,1,1,0.0839695,1,-0.806452,0,-0.333333,
    
26.   0.708333,1,1,-0.320755,-0.105023,-1,1,-0.419847,-1,-0.225806,0,1,
    
27.   0.583333,-1,0.333333,-0.603774,1,-1,1,0.358779,-1,-0.483871,0,-1,
    
28.   0.166667,1,-0.333333,-0.433962,-0.383562,-1,-1,0.0687023,-1,-0.903226,-1,-1,
    
29.   0.458333,1,1,-0.358491,-0.374429,-1,-1,-0.480916,1,-0.935484,0,-0.333333,
    
30.   0.875,-1,-0.333333,-0.509434,-0.347032,-1,1,-0.236641,1,-0.935484,-1,-0.333333,
    
31.   0.5,1,1,-0.509434,-0.767123,-1,-1,0.0534351,-1,-0.870968,-1,-1,
    
32.   0.125,1,0.333333,-0.320755,-0.406393,1,1,0.0839695,1,-0.806452,0,-0.333333,
    
33.   0.25,1,1,-0.698113,-0.484018,-1,1,0.0839695,1,-0.612903,0,-0.333333,
    
34.   0.291667,1,1,-0.132075,-0.237443,-1,1,0.51145,-1,-0.612903,0,0.333333,
    
35.   0.416667,-1,1,0.0566038,0.283105,-1,1,0.267176,-1,0.290323,0,1
    
36. };
    
37. 
    
38. 
    
39. int main()
    
40. {
    
41.   Mat trainData(10, 12, CV_32FC1);//构建训练样本的特征向量
    
42.   for (int i=0; i<10; i++)
    
43.   {
    
44.     for (int j=0; j<12; j++)
    
45.     {
    
46.       trainData.at<float>(i, j) = inputArr[i][j+1];
    
47.     }
    
48.   }
    
49. 
    
50.   Mat trainResponse(10, 1, CV_32FC1);//构建训练样本的类别标签
    
51.   for (int i=0; i<10; i++)
    
52.   {
    
53.     trainResponse.at<float>(i, 0) = inputArr[i][0];
    
54.   }
    
55. 
    
56.   CvNormalBayesClassifier nbc;
    
57.   bool trainFlag = nbc.train(trainData, trainResponse);//进行贝叶斯分类器训练
    
58.   if (trainFlag)
    
59.   {
    
60.     cout<<"train over..."<<endl;
    
61.     nbc.save("d:/normalBayes.txt");
    
62.   }
    
63.   else
    
64.   {
    
65.     cout<<"train error..."<<endl;
    
66.     system("pause");
    
67.     exit(-1);
    
68.   }
    
69. 
    
70. 
    
71.   CvNormalBayesClassifier testNbc;
    
72.   testNbc.load("d:/normalBayes.txt");
    
73. 
    
74.   Mat testSample(1, 12, CV_32FC1);//构建测试样本
    
75.   for (int i=0; i<11; i++)
    
76.   {
    
77.     for (int j=0; j<12; j++)
    
78.     {
    
79.       testSample.at<float>(0, j) = testArr[i][j];
    
80.     }
    
81.     float flag = testNbc.predict(testSample);//进行测试
    
82.     cout<<"flag = "<<flag<<endl;
    
83.   }
    
84. 
    
85. 
    
86. 
    
87.   system("pause");
    
88.   return 0;
    
89. }
    

复制代码

一只人

smallyouyou 发表于 2016-5-15 22:39
用類神經做看看，
這邊有基本的用法，
我還沒用過opencv這塊，

谢谢提供信息，我看看。

piaobuliao

hsv好像还YCrCb的区分度好

一只人

最后用了svm效果相当好。谢谢各位了

xywhnh

SVM是什么东东啊楼主

13182196109

xywhnh 发表于 2016-7-2 07:51
SVM是什么东东啊楼主

支持向量机算法，比较经典了

lxxxl

请问一下，你说的用SVM识别颜色的具体实现步骤是怎样的啊？我也用HSV实现了颜色的识别，但是效果也是不是很理想

爱笑的十三

请问对于颜色的识别而言有没有什么识别更好的算法呢