Robust location recognition based on efficient feature integration

We present an effective location recognition approach by integrating multiple visual cues. We aims at developing a localization system with high accuracy and efficiency. Although one kind of features such as interest points might be useful in certain environment, multiple cues are important for robust location recognition. We use three kinds of features in this work: edges, interest points, and color features. Vision-based localization consists of two online steps: feature detection and feature matching. The efficiency of a localization system depends on the computational complexity of the two steps. To accelerate feature detection, the three kinds of features are detected in an integrated process: edges are detected directly based on the Harris function; Harris interest points are detected at several scales and Harris-Laplace interest points are found using the Laplace function. Corner regions are characterized by the SIFT descriptor. Color histograms are used to describe edge and smooth regions. The color histograms are refined by using the color coherence method to include spatial information. The integration of multi-cue brings efficient feature detection and high recognition ratio to the localization system. Our algorithm can be extended to recognize images containing specific objects, scenes or buildings. Experimental results show the proposed approach has good performance.