Human Motion Tracking And Analysis In Multi-camera Environment

Human motion tracking and analysis is a hot and open research topic in the past decades for its wide application areas. Despite the fact that good results were achieved by traditional approaches, there still exist some limitations in real-world scenarios. This dissertation focuses on geometrical and spatial constraints among multiple cameras, proposed novel human motion tracking algorithms. The human motion is analysed in several levels based on semantic structure, while a context-aware framework and related models, with the goal of selectivity and task-oriented procedure, are built to fuse human motion information from different levels. The contributions are listed as follows:Firstly, a fast and robust person localization algorithm in multi-camera environment is proposed by utilizing the geometrical concurrent constraints regarding perpendicular lines in multi-camera environment. This algorithm runs in the condition of neither full calibration of cameras nor assumption on the visibility of particular body part. Real-world experiments have demonstrated robustness, accuracy and computational efficiency of this algorithm.Secondly, by using the ?-shaped curve of human head and shoulder as the tracking target, a novel tracking method is proposed to integrate tracking and detection into a unified model, where human posture, location and head-shoulder contour can be obtained simultaneously. The robustness and efficacy of the approach have been confirmed by challenging real-world scenarios.Thirdly, a context-aware framework for human motion tracking and analysis is developed. Motion information is used to drive the inference of motion scenario in the proposed context model for a coarse-to-fine procedure of human analysis. The experiments in activities of daily living prove that context plays an important role in human motion tracking and analysis in real-world scenarios by lowing down system load and improving accuracy and robustness.Fourthly, a generic framework for distributed multi-stream system under context-aware environment is built to fulfill the common requirements among context-aware systems. The proposed framework can not only facilitate the construction and deployment of context-aware multi-stream systems, but also manage context information in an effective way, providing excellent extendibility.