보행자 인식을 위한 Interpolation 기법에 따른 인터리빙 HOG의 성능 분석

This paper presents the comparison of interpolation methods on Bin-interleaved Histogram of Oriented Gradients (Bi-HOG). The feature dimension of our Bi-HOG is a half size of HOG using bin-interleaved method. We experimentally demonstrate that SVM classifiers trained by Bi-HOG have the almost same detection performance as one by the original HOG. Especially 3D interpolation method has the best performance on the both HOG and Bi-HOG and Bi-HOG in 3D-interpolation hardly has performance reduction in comparison with the original HOG. And Bi-HOG considerably reduce storage requirement and simplify the computational complexity about 55%.