Person Detection and Tracking
Detecting and tracking human beings in a given scene represents one of the most important and challenging tasks in computer vision, nowadays far from being solved. We are interested in these issues for their implications in video-surveillance and driver assistance systems.
People detection is challenging because of the high variability of the human body appearance (different poses, clothes, viewpoints, etc.). Detecting a person in crowded scenarios (which are common in video-surveillance applications) is even harder, due to the severe occlusions that may occur. Our current research focuses on discriminatively learning a set of spatial relations between Poselet types using Random Forests.
We are also interested in computational issues, which are usually neglected by the computer vision community mainly focusing on accuracy problems. In this context, Embedded Computer Vision plays a fundamental role. Our goal is to design and implement vision algorithms for high-performance dedicated hardware platforms (e.g., FPGA, GPU).
Tracking can be seen as an extension of object/people detection over time, generating the trajectory of each object/person through the video. Here we focus on issues such as people appearance change over time and occlusions. Occlusion problems are particularly hard to deal with when the occluding and the occluded persons have a similar appearance.
- S. Martelli, D. Tosato, M. Cristani, V. Murino,
"FPGA-based pedestrian detection using array of covariance features"
5th ACM/IEEE International Conference on Distributed Smart Cameras (ICDSC), 2011