Human Detection using HOG Feature

Steps

• Read image. Convert to grayscale using: Gray = round(0.299Red + 0.587Green + 0.114*Blue)
• Gradient calculation using prewitt’s operator, magnitude calculation (sqrt(Gx^2+Gy^2)) and normalization.
• Gradient angle calculation.
• Calculate histogram bins for every cell (unsigned format/9 bins).
• Calculate normalized bins for every block (l2 norm).
• Flatten and concatenate normalized bins for every block to get a descriptor of length 7524.
• 3NN implementations using similarity formula: sum(min(input, train))/sum(train)

Here is the link to the code.