Segmentation and Object Recognition in Road Scenes Using Multi-scale Semantic Bag of Textons Method

摘要

While the bag of words models are popular and powerful method for generic object recognition, they completely discard the spatial layout and context information. This paper presents a novel method for segmentation and object recognition of road environment scenes using multi-scale semantic bag of textons method. The histogram of extracted texton is concatenated to regions of interest with multi-scale regular grids. This feature can learn automatically spatial layout and relative positions between object and object in a road scene. Experimental results show that the proposed method can effectively classify the texture-based objects of a road scene comparing the conventional bag of texton methods. In the future, the proposed system can combine with a scene understanding system for vehicle environment perception.