Lightweight video object recognition based on sensor fusion

While there are several promising approaches for visual object recognition the application of lightweight devices under varying image conditions and using low quality images still causes lots of problems to be solved. We introduce a new retrieval mechanism including standard orientation sensors helping the visual recognition process. We apply a view based model of the objects and the matching of the query and candidate images is based on compact image descriptors coupled with relative orientation. Besides introducing the concept we show the effectiveness of our approach through two datasets using noisy and blurred images.