Robust object detection and localization for real-time autonomous surveillance applications

In this dissertation, we present robust object detection and localization methods for real-time autonomous surveillance applications. The hyperspectral image processing technology can provide high performance object detection, while requiring special purpose camera and high computational complexity. For real-time processing, the complexity reduction is critical. On the other hand, conventional detection methods can be easily deployed in surveillance system using general purpose camera, although it suffers from limited detection performance. We propose a human body and face joint detection method with multiple cameras where the detection performance is significantly improved. We also propose an object localization method which collaborates with the detection method so that the overall performance can be improved.;Next, a simplified algorithm to localize object position using multiple images is proposed. We use a parallel projection model which supports both zooming and panning capabilities of the imaging devices. Our proposed algorithm is based on a virtual viewable plane for creating a relationship between an object position and a reference coordinate. The reference point is obtained from a rough estimate which may be obtained from the pre-estimation process. The algorithm minimizes localization error through the iterative process with relatively low computational complexity. In addition, non-linearity distortion of the digital image devices is compensated during the iterative process. The performances of several scenarios are evaluated and analyzed in both indoor and outdoor environments.;Finally, we propose a human body and face joint detection method in a multiple camera environment. The limitations of single camera based human detection are addressed. Through the multiple cameras, the observable range becomes broader with additional perspectives. Multiple cameras with different perspectives pave the way to collaborate one another, and enable to support additional information among cameras. Each detected human from multiple cameras is transferred to a global localization, which enables us to monitor all-around human movement in a global coordinate. The global information reversely assists the original detection which suffers from the single camera limitations. Furthermore, our proposed application supports the camera panning and zooming through the global information. The performances of multiple human detections are evaluated and analyzed in a variety of multiple camera environments.;First, the spectral characterization for efficient image detection using hyperspectral processing techniques is presented. We investigate the relationship between the number of used bands and the performance of the detection process in order to find the optimal number of bands. The band reduction significantly reduces computation and implementation complexity of the algorithm. Specifically, we define and characterize the contribution coefficient for each band. Based on the coefficients, we heuristically select the required minimum bands for the detection process. We have shown that the small number of bands are efficient for effective detection. The proposed algorithm is suitable for low complexity and real-time applications.