The Histograms of Oriented Gradients (HOG) based on the luminance gradient and are often used for person detection from images. Based on a similarity between luminance gradients and frequency components of image, we propose to use the fast Fourier transform (FFT) for person detection. First we apply FFT to a block divided image. Amplitude information and phase information are calculated, then those values are normalized at each block. All the normalized values of the image are defined as FFT-based feature. It is shown that the detection accuracy of