Detecting and tracking moving objects from a moving platform

Detecting and tracking moving objects are important topics in computer vision research. Classical detecting and tracking methods for steady cameras are not suitable for use with moving cameras because the assumptions for these two applications are different. This thesis aims to develop algorithms that can detect and track moving objects with a non-fixed position camera. To achieve this aim, we analyze the image sequences captured by a moving camera undergoing general 3D rotation and translation. New computer vision algorithms are developed to obtain feasible solutions to the problem without prior camera calibration and classifier training.;The initial step of this research is to develop a new method for estimating camera motion parameters. Based on the estimated camera motion parameters, two methods are developed for detecting moving objects: one based on the Bayesian decision and another based on the belief propagation. The Bayesian decision method uses camera motion parameters to compensate for the camera motion. The background classification rule for every pixel is developed to generate a foreground mask, and then the moving objects can be detected. Another detection method addresses the detection problem by creating a graphical model, which uses the belief propagation algorithm. After camera motion parameters are estimated, feature points in every frame are grouped using a hierarchical clustering algorithm. Then, the related groups between adjacent frames are linked, which results in a graphical model. A belief propagation algorithm is used to transmit the information on this graphical model to find which group is on the moving object. x.