Moving Person Detection Using Blob Descriptors and Support Vector Machine

In this paper a method dedicated to people detection in videos is presented. We have constructed a system which extracts moving objects using background subtraction, then computes HOG descriptors for each object and classifies the objects into “person” or “non-person” using a Support Vector Machine. We have also analyzed possible improvements for background model updating during background subtraction and the opportunity of using a simplified, faster than HOG, descriptor for moving objects. We present and comment our experimental results, pointing out advantages and disadvantages.