Probabilistic framework for multi-target tracking using multi-camera: applied to fall detection

摘要: The developments in health care lead to longer life expectancy developed countries. growth of the number seniors put new challenges services, caregivers and family members. One major is how prevent elderly people from falling down. Especially, if person lives alone, can be quite dangerous for his / her health. Falling down cause serious consequences like broken bones or lost consciousness people. Therefore, it key importance able provide help as soon possible, or, even try incident. Combining video-based surveillance computer vision techniques could bring us towards intelligent systems that monitor raise an alarm case unusual events. main question automate fall detection using only video information (like camera network) with application public services (i.e., houses, hospitals). Generally, when talking about systems, there are some elements would remain same every system irrespective application. These are: tracking, data association context network post-processing based on extracted activity recognition). So, this thesis addresses each these questions tries find a generalized framework integrate all methods one system. criterion project was avoid calibration is, correspondence between views should confidence probability function views. decided adopt Bayesian multi-target multicamera tracking [1] fulfills our requirements. When considering single view camera, allows represent target set samples (or particles) have probabilistic measures (weights) proportional likelihood being at position. For multi-camera system, needs associate corresponding targets among Gale-Shapley algorithm finding stable matching two classes (in cameras) modified purpose. part, silhouetteand velocity-based features were passed through Support Vector Machine (SVM) separate events other activities. A test videos recorded cameras stability reliability algorithm. initial results encouraging. Further integration additional may improve performance