Current system has 3 problems: 1) It doesn’t account for camera motion; 2) The moving objects are not separated from the background; 3) The motion feature is too simple. Many motion properties can be extracted in the motion feature. This project aims to build a more robust and detailed feature. It follows 3 steps: 1) Apply keypoint matching method to eliminating camera motion; 2) Build GMM background model to detect moving objects and segment them; 3) Train a SVM classifier to recognize objects and extend motion feature with object numbers, sizes and colors, etc.

Organization

Student

zydu

Mentors

  • Claudiu Tanase
close

2016