Real-time human detection based on cascade frame

A real-time pedestrian detection approach with two steps is proposed in this paper. The first step is the detection by HOG in combination with the classifier of cascade frame. The weak classifer in cascade is Boosting which corresponds to block features of HOG. To make it more accurate in feature selection we define a model of feature selection to limit the range of feature block to the edge of human in detect window. The second step is to extract the head image in positive window and compute the color histograms as feature. Traditional AdaBoost is used to validate the detection result. Only when a window passes both steps it is judged as a human. The experiment result in the paper shows that the approach is effective and real-time detection is implemented.