In surveillance and security applications, the use of vision and infrared sensors has become increasingly prevalent. These sensors play a crucial role in monitoring critical infrastructures such as harbors, airports, and military camps. To maximize their efficiency and performance, automatic processing of sensor data and intelligent control of the sensor is essential. This thesis focuses on two important aspects of the sensor fusion system: target tracking and sensor control.
One of the key challenges in target tracking is dealing with road-constrained targets tracked by an airborne video or infrared camera. By leveraging road network maps, the thesis proposes a dynamic model for on-road target tracking with a camera. To handle the computational load, a Rao-Blackwellized particle filter is employed. Furthermore, a pedestrian tracking framework is developed and evaluated through a real-world experiment. The integration of contextual information, such as road network data, not only enhances tracking performance but also facilitates track analysis, anomaly detection, and efficient sensor management.
The thesis also delves into planning for surveillance and reconnaissance, which encompasses various problem definitions and applications. Two specific problems are explored. The first problem involves multi-target search and tracking, where the goal is to control the trajectory of an aerial sensor platform and its camera’s pointing direction to track identified targets while searching for new ones. To facilitate effective planning, the thesis proposes a measure that enables the comparison of different tracking and searching tasks within a unified framework. An algorithm based on this measure is developed, and simulation results in an urban area scenario are provided.
The second problem addressed is aerial information exploration for single target estimation and area surveillance. In the case of a single target, the objective is to optimize the trajectory of a sensor platform with a vision or infrared camera to enhance target estimation performance. This problem is approached from both an information filtering and particle filtering perspective. In the context of area exploration, the task involves capturing useful image data of the designated area by controlling the sensor platform’s trajectory and camera pointing direction. Successful exploration entails multiple images from different viewpoints. The thesis proposes a method based on multiple information filters and presents simulation results from area and road exploration scenarios.
Overall, this thesis emphasizes the significance of efficient target tracking, sensor control, and planning in surveillance and security applications. By leveraging innovative techniques and incorporating contextual information, the aim is to enhance surveillance effectiveness, anomaly detection capabilities, and sensor management efficiency. The findings presented in this thesis provide valuable insights and contribute to the ongoing efforts in advancing surveillance technology for the protection of critical infrastructures and public safety.
