Histogram of Oriented Gradients-based Human Detector Desktop Application (using OpenCV trained SVM model)
This application shows how the HOG-based Human Detector works step by step:
- Build Image Pyramid
- Slide detection windows at every position in every scaled image
- Compute HOG feature for every detection windows
- Multiply the HOG feature of detection windows with SVM model to compute confidence score
- Perform Non-Maximum Suppression to pick the highest scoring box in a local location (this functionality is pending)
I use the Square Root Approximation (SRA) technique in this paper to calculate the HOG feature. I also use the proposed Gradient Vote method in that paper.
I utilize multiple threads when processing multi-scale images in Image Pyramid to handle them simultaneously.
The purpose of this project is for understanding how to extract HOG feature step by step.
Use at your own risk.