Statistical Image Segmentation and Region Classification Approaches for Automotive Radar

This paper discusses statistical approaches currently being investigated to perform image segmentation and region classification in high resolution automotive radar imagery in the complex urban environment. The purpose is to identify the free traversable space ahead of the vehicle which would ultimately provide input into autonomous vehicle path planning algorithms. Three general methodologies are described, all based on the image region pixel intensity statistics which vary depending on the features within the imaged scene. Results show the promising potential for segmentation with these methods and some examples of segmentation and identification have been demonstrated for chosen road scene region types of asphalt, grass, shadow and objects/other.