Image Based Localization Using Semantic Segmentation for Autonomous Driving

One of the research topics that maintains its popularity in the field of Computer Vision is place recognition and localization for autonomous vehicles. It is a known fact that GPS systems used for localizing vehicle cannot be activated in some cases and this inability has accelerated image based positioning studies. In our study, we performed image based localization using dataset that includes Malaga city center images. Firstly, a semantic descriptor is obtained as a result of semantic segmentation and localization was performed using the approximate nearest neighbor search. After that, success of this method was compared with the success of the local descriptor based method which is frequently used in the literature. Furthermore, a hybrid method obtained by combining these two methods is proposed. The superiority of the proposed hybrid image-based localization method, and hereby contribution of the semantic descriptor is demonstrated by experimental results.