In order to build an effective machine learning solution, you will need the proper analytical tools for evaluating the performance of your system. Cross-validation is one of the most important tools, as it gives you an honest assessment of the true accuracy of your system.
To help in my understanding of the HOG descriptor, as well as to allow me to easily test out modifications to the descriptor, I wrote functions in Octave / Matlab for computing the HOG descriptor for a detection window.
One of the most popular and successful “person detectors” out there right now is the HOG with SVM approach. When I attended the Embedded Vision Summit in April 2013, it was the most common algorithm I heard associated with person detection.
Gradient vectors (or “image gradients”) are one of the most fundamental concepts in computer vision; many vision algorithms involve computing gradient vectors for each pixel in an image.
I found it really hard to get a basic understanding of Support Vector Machines. To learn how SVMs work, I ultimately went through Andrew Ng’s Machine Learning course (available freely from Stanford). I think the reason SVM tutorials are so challenging is that training an SVM is a complex optimization problem, which requires a lot of math and theory to explain.