Online tracking of multiple objects using WiSARD

Abstract : This paper evaluates the WiSARD weightless model as a classification system on the problem of tracking multiple objects in real- time. Exploring the structure of this model, the proposed solution applies a re-learning stage in order to avoid interferences caused by background noise or variations in the target shape. Once the tracker finds a target at the first time, it applies only local searches around the neighborhood in order to have fast response. This approach is evaluated through some experiments on real-world video data.