“…In the state-of-the-art approaches, most researchers have been working on two issues: feature extraction methods (handcrafted features and deep convolutional features) and classification through the machine learning algorithms. Some of the promising handcrafted feature extraction methods are the histogram of oriented gradients (HOG) [4-6, 14, 24, 25], Haar wavelet [2,[25][26][27], scale-invariant feature transform (SIFT) [28][29][30], edge templates [5,23,31,32], adaptive contour features [2,23,33,34], Gabor filters [15,27], covariance descriptors [11,15,19,35], and local binary pattern (LBP) [6,11,24,36]. Among these handcrafted features, the histogram of oriented gradients (HOG) is a well-known feature descriptor for pedestrian detection due to the rich feature information under different illumination changes [14,16,20].…”