Model modification methodology for adaptive object recognition in a sequence of images
Sung Wook Baik, Peter W. Pachowicz · 1999
This dissertation is focused on the development and experimental validation of an on-line model modification methodology for the recognition of objects. The methodology deals with the problem of continuously changing object characteristics used by the recognition processes. This characteristics change is due to the dynamics of scene-viewer relationship such as resolution and lighting. The approach applies a modified Radial-Basis Function paradigm for model-based object modeling and recognition. On-line adaptation of these models is developed and works in a closed loop with the object recognition system to perceive discrepancies between object models and varying object characteristics. Feedback re-inforcement mechanism defines and activates selected model modification behaviors to adjust local elements of the object model. These behaviors include component accommodation by parameter adjustment, component translation over the feature, space, new component generation, and extinction of unused components. Reinforcement learning transforms detected confidence change from the data and model matching into parametric adjustments applicable to the appropriate model components. Developed methodology is experimentally validated through several experiments and using developed fully autonomous model evolution system. First, preliminary experiments are run to justify the development of model modification mechanism and its application to the on-line object recognition in dynamic environments. These experiments are shown on a sequence of images (a composition of textures in the scene) taken under gradually changing resolution and lighting conditions. Next, developed methodology and a prototype system are tested through three experiments, where model modification behaviors are gradually integrated into a complete system. Finally, the experiments are completed for a sequence of outdoor images.