Efficient pedestrian detection by Bin-interleaved Histogram of Oriented Gradients
Haengseon Son, Seon-Young Lee, Jongchan Choi, Kyung‐Won Min · 2010
This paper presents an efficient pedestrian detection by Bin-interleaved Histogram of Oriented Gradients (Bi-HOG) for automotive applications. The state-of-art feature named HOG [5] is adopted as the basic feature. We arrange alternately even-bin cells and odd-bin cells in one block and then extract the only even-bin feature elements for even-bin cells and the only odd-bin feature elements for odd-bin cells. So the feature dimension of our Bi-HOG is a half size of HOG by bin-interleaved method like this. We experimentally demonstrate that SVM classifiers trained by Bi-HOG have the same detection performance on the DaimlerChrysler data set as one by the original HOG in our two-staged pedestrian detection system and considerably reduce storage requirement and simplify the computational complexity.