Can anyone point me to any examples or tutorials on how to use random trees in OpenCV, other than the opencv documentation. I would like to use it for hand detection in images. The idea is to use Local Binary Patterns to extract features and a Random Forest to classify whether a hand is detected or not.
It will be used along with skin detection to reduce areas that are not a hand.
Any advice will be highly appreciated.
here is, what i used recently for the random forest:
Mat train_features; // 1 row for each lbp feature, float !
Mat train_labels // 1 row (containing 1 float element with the id) for each feature
CvRTrees tree;
CvRTParams cvrtp; // set params, play with those!! (the numbers were for my problem)
cvrtp.max_depth = 25;
cvrtp.min_sample_count = 6;
cvrtp.max_categories = 2;
cvrtp.term_crit.max_iter = 100;
tree.train ( train_features , CV_ROW_SAMPLE , train_labels,cv::Mat(),cv::Mat(),cv::Mat(),cv::Mat(),cvrtp );
// later:
float id = tree.predict(sample); // where sample is a lbp feature , again
but again, since your problem is more a binary classification, not a multi-class one, another classifier, like svm or dtree might give better results.
if you look at the lbp-FaceRecognizer , it's not even using any of those, just a plain chi-square based knn. (and seemingly, results did not improve using svm or such)
please have a look here, for a nice tut on machinelearning in opencv
@paioye, with questions on LBP and Random Forests, please stick to your new question. Also accept this answer if it helps you out for the Random Trees part.
Asked: 2013-03-11 07:00:25 -0600
Seen: 1,001 times
Last updated: Sep 01 '13
