| An articulated object can be loosely defined as a structure composed of links and joints. The human body is a good example of a nonrigid, articulated object. Localizing body shapes in still images remains a fundamental problem in computer vision, with potential applications in surveillance, video annotation, HCI, and entertainment. This thesis explores a 2D model-based approach to this problem.; We start from a fixed viewpoint scenario (side-view walking) by introducing a landmark-based model of the body contours. Several image cues are combined in a Bayesian frame-work, including edge gradient, silhouette, skin color, and region similarity. The model is arranged into a sequential structure, enabling simple yet effective spatial inference through Sequential Monte Carlo (SMC).; Next, we extend the system to situations where the viewpoint of the human target is unknown. To accommodate large viewpoint changes, a mixture of view-dependent models is employed. Each model is decomposed based on parts, with anthropometric constraints and self-occlusion explicitly treated. Inference is done again by SMC, searching in parallel with dynamic resource allocation.; Finally, we study body localization in a generic setting: single image, arbitrary pose, and arbitrary viewpoint. The proposed solution is a hybrid search facilitated by a 3-level hierarchical decomposition of the model. We first fit a simple tree-structured model by Dynamic Programming (DP). The output is a series of proposal maps that encode the probabilities of partial body configurations. Next, we fit a mixture of view-dependent models by SMC, which handles self-occlusion and large viewpoint changes. DP and SMC search in opposite directions such that the DP proposals are utilized effectively to initialize and guide the SMC inference. This combination of deterministic and stochastic search ensures both the robustness and efficiency of DP, and the accuracy of SMC. Finally, we fit an expanded mixture model with increased landmark density through local optimization.; The models were trained on a large number of gait images. Extensive tests on cluttered images with varying poses including walking, dancing and various types of sports activities demonstrate the feasibility of the proposed approach. |
