CvSVM - decision function value
Decision function value - that is signed distance to the margin
But, if I have 2 class classification with linear kernel, how can I figure out on which side of the hyperplane am I?
When you define your training samples as +1 and -1 for positive and negative samples, the value > 0 will indicate a positive classification and a value < 0 will indicate a negative classification.
I think so, but when i run the program image by image, when there no human the decision function value is around 0.1-0.8 (with different images) and when there is human is from -0,001 to 0,009 .... 0_o
Thats because humans are difficult class to simple classify binary using an svm ... use more examples to refine the margin. Also, the -0,0001 are just misclassification errrors, which will always happen. You should read more documentation on the SVM technique.