Road Scene Image Segmentation Based on Feature Fusion

Image segmentation in machine vision is a significant step to get a holistic image understanding of the captured scene. Region-based, edge-based, clustering, learning-based methods are image segmentation approaches and there are plenty of methods for each one. Beside these methods, image fusion as a branch of data fusion is related to machine vision by enhancing image in different ways and levels (pixel, feature and decision). In autonomous vehicles (AV) and Advanced assistant driver systems (ADAS) cameras are important sensors those widely used to control and navigate system. Front view camera gives an image from frontal scene containing different objects that should be processed during driving phase. Making a good segmentation from the scene helps AVs or ADAS systems to tackle their duties. In this research, we have used fusing different segmentation and object Classification in order to get a better perception of surrounding area. On the other hand, the system should complete all process in fraction of a second, so it is so important to reach the best execution time for AVs' real time applications. K-Mean and Fuzzy C-Mean (FCM) are clustering methods and also two methods for fusing different specific parts (features) of image to a class or fused decision in high level data fusion. There for, we will reach to a segmented image in which background and foreground are completely separated.