Reliable and efficient tracking of human motion using particle filtering

This dissertation presents three novel methods for improving the efficiency and accuracy of human tracking using particle filtering.;The first method exploits multi-modal visual data from both Electro-optical (EO) and Infrared (IR) cameras for human object tracking. Particle filtering is used to perform object tracking as well as multi-modal data fusion. A centroid-based technique is introduced to discover potentially moving human objects and obtain related coordinates. Once moving targets are located, both EO and IR features are combined to extract object templates for spreading particles. To determine and update the importance of each particle, statistic information of a blob centered at the current particle is compared with available templates. Consequently, particles making greater contribution to predict state space changes are assigned with higher weights. The simulation results show that robust human tracking can be achieved through joint EO and IR data processing.;The second method incorporates Gaussian Process Dynamical Model (GPDM) to improve the efficiency in particle filtering based multi-target tracking. With the proposed Particle Filter Gaussian Process Dynamical Model (PFGPDM), a high dimensional target trajectory dataset of the observation space is projected to a low dimensional latent space in a nonlinear probabilistic manner, which will then be used to classify object trajectories, predict the next motion state, and provide Gaussian process dynamical samples for the particle filter. In addition, histogram-Bhartacharyya, GMM-Kullback-Leibler, and rotation invariant appearance models are employed, and compared in the particle filtering as the complimentary features to coordinate data used in GPDM. The simulation results demonstrate that this approach can track more than four targets with reasonable run-time overhead and performance. In addition, it can successfully deal with occasional missing frames and temporary occlusions.;The third method employs an interleaved object detection and tracking approach to improve the performance of PFGPDM with unreliable initial detections. An online AdaBoost learning algorithm is used in the detection of human bodies and faces. The detection results, i.e. the coordinates of the centroid or the left corner of an object are provided to the tracking system. The simulation results indicate that the proposed frame can effectively detect and track new objects entering the scene.