| Summary: | As the amount of video data grows larger every day, the efforts to create intelligent systems able to perceive, understand and extrapolate useful information from this data grow larger, namely object detection and tracking systems have been a widely researched area in the past few years. In the present work we develop a real time, multiple camera, multiple person detection and tracking system prototype, using static, overlapped, sh-eye top view cameras. The goal is to create a system able to intelligently and automatically extrapolate object trajectories from surveillance footage. To solve these problems we employ different types of techniques, namely a combination of the representational power of deep neural networks, which have been yielding outstanding results in computer vision problems over the last few years, and more classical, already established object tracking algorithms in order to represent and track the target objects. In particular, we split the problem in two sub-problems: single camera multiple object tracking and multiple camera multiple object tracking, which we tackle in a modular manner. Our long-term motivation is to deploy this system in a commercial application, such as commercial areas or airports, so that we can build upon intelligent visual surveillance systems.
|
|---|