Human detection method based on feature co-occurrence of HLAC and HOG

This paper proposes a new human detection method, which is robust to illumination change and does almost not confuse human with other objects even they have similar contours. This method is based on integration with two features: Higher-order Local Auto-Correlation (HLAC) features and Histograms of Oriented Gradients (HOG) features. HLAC features can give a broad pattern of gray scale image. The features are invariant to shift, but poor at illumination change. HOG features can give an accurate description of contour of human body. These are robust to illumination change, but usually confuse human with other thing that similar contour. In order to make up these deficits, we use co-occurrence of multiple features to integrate HOG and HLAC features. In our experiments, we obtain 20.1% lower value on false positive per window than other proposed method when miss rates are similar. These results proved the effectiveness of this new method.