Real-Time and Event-Triggered Object Detection, Recognition, and Tracking

This paper presents the current state of a novel event-based surveillance framework for real-time detection and tracking of the person-of-interest with IP PTZ network camera. Formulating the problem in a (non-linear) Bayesian filtering framework in combination with Convolutional Neural Networks (CNN), we develop dynamical and adaptive approaches for identifying the Person-of-Interest(PoI) from its given signature. We identify the target as human or non-human then recognize the facial Histogram of Oriented Gradients (HOG) feature of the PoI to finally start tracking the PoI across the surveillance area covered by the IP PTZ network camera. In particular, this research work intends to incorporate and combine recently developed event-based tracking algorithms with deep learning algorithms and implement in a practical setting for real-time detection and tracking via a IP PTZ network camera.