Trained with different SVM kernels. SVM parameters affects the overall detection performance.
SVM also consumes much more processing time which I can do nothing to optimize so far.
The following are the results of different SVM parameter settings.
According to the reference paper, the group trained for a few days. It should be understood that usually the more training data the better.
As the deadline is approaching, it seems impossible for me to apply non-maximum suppression, instead I’ll select best performance SVM parameter combination ,run error check and get all benchmark ready.