Making Sense: Robust Approaches to Visual Monitoring
The main focus of this research project is the development and improvement of techniques of computer vision and machine learning. The purpose of these techniques is the monitoring of environments by using visual data obtained through a network of surveillance cameras. One of the main objectives of automatic environment monitoring is the extraction of information about human activities to detect interactions between agents and to identify patterns of suspicious behavior. For the activities to be analyzed, a set of problems, such as detection and identification of agents in the scene, tracking over time (possibly between distinct cameras) and recognition of individual actions, needs to be solved. Such problems comprise the domain of applications called people observation, responsible for the analysis of images and videos containing humans. If on the one hand surveillance cameras provide a large amount of visual data, on the other hand, the analysis of this data becomes a great challenge when done manually. Therefore, the understanding and automatic interpretation of activities carried out by humans in videos is of great interest to assist in the tasks of security agents.
Visual surveillance is responsible for the real-time and automatic monitoring of the agents present in an environment being imaged by a network of cameras. Its main objectives are: to provide automatic interpretation for the scene and the automatic understanding of the activities and interactions between the agents being observed from large amounts of visual data. This project aims to develop approaches to solve problems related to visual surveillance. It focuses on the aspects of knowledge extraction from surveillance scenarios, including the problems of activity recognition, detection of anomalous events, identity maintenance and optimization of deep neural networks. The solutions presented to the aforementioned problems, aside from advancing the state of the art, can be incorporated into surveillance systems. This results in technological advances with social impact, since such systems provide safer environments for the circulation of people. It also has economical impact, due to the great market potential for systems of this nature in our country.