Vision-based object recognition under atmospheric distortions

Pui Anantrasirichai

The presence of heat haze and atmospheric turbulence can significantly reducing the effectiveness of conventional approaches to target classification and tracking targets in surveillance imagery. This, in turn, makes scene interpretation and analysis extremely difficult, and hence impacts situational awareness. This project will address this challenge using modern data driven machine learning methods. The main then objectives are: i) to develop real-time video restoration techniques that mitigate spatio-temporal distortions and thus improve visual interpretation of the scene by a human observer, and ii) to assist decision making by implementing and evaluating real-time object recognition and tracking using the restored video.