DOAS 2008 project Looking-at-people: intelligent surveillance systems

Intelligent Autonomous Systems increasingly often operate in environments inhabited by humans, like houses, public places (stations, shopping centers) or intelligent vehicles. In order to properly interact with people an intelligent system has to detect people in its environment, identify them and possibly recognize their behavior. Examples of such systems are intelligent security systems that detect people entering restricted areas or detect aggressive behavior, intelligent vehicles that detect pedestrians (and warn the driver if a pedestrian is to close to the vehicle) or interface robots that act as an interface for taking user’s commands and presenting results. Perception of humans and their actions typically relies on cameras as sensors which capture various details about the location, number and appearance of people. Therefore computer vision techniques for “looking-at-people” are the mainstream of research in Intelligent Autonomous Systems. The relevant algorithms range from low-level image processing techniques (edge detection, pixel classification) to higher-level gesture and object classification methods. The DOAS 2008 project will be performed in the context of larger research at the ISLA laboratory which develops computer vision techniques for “looking-at-people”. Currently, the laboratory runs several projects that aim at detecting people in still images, tracking people through images sequences, detecting at tracking individual body parts or pose estimation of humans in 2D and 3D. In our project of interest, CASSANDRA [3], the final goal is to automatically detect aggressive behavior of people in public spaces like train stations. The laboratory provides an opportunity for students to get hands-on experience with several advanced image processing tools. Additionally, there are several realistic video sequences involving professional actors recorded in real-world setup (train station). This dataset will be used in the project to evaluate the developed algorithms.