Local probability based safe region detection for autonomous driving

Abstract

This paper proposes a new approach to detect the driving region and to detect the driving possible region from image sequence. To achieve this, we use local adaptive threshold and local probability for detecting the driving region and for detecting the driving possible region, respectively. Here are the three main aspects. The first one is the driving region detection. For this we use the local adaptive threshold. The second one is to recognize the driving possible region. To do this, we use a randomly selected initial seed and its extension using the distance between local probabilities. The third one is to combine the driving and the driving possible regions. It gives better results for safe autonomous driving. Sometimes, the driving region is not detected correctly due to very great noise factors. In this case the possible driving region still helps autonomous driving.