Visual Scene Understanding for Self-Driving Cars Using Deep Learning and Stereovision

2019-02-07T16:16:05Z (GMT) by Amélie Grenier
<div>Poster presented at the 2018 Defence and Security Doctoral Symposium.</div><div><br></div><div>Autonomous driving has been rapidly evolving for the last few years and there is a lot of fervour in increasing the intelligence of these vehicles. One key aspect of a self-driving car is its ability to sense the environment in order to be aware of its surrounding.</div><div>Our interest lies in using computer vision and deep learning techniques to detect surrounding entities; localising and recognising them. Here, we present a novel deconvolutional neural network for semantic segmentation, combined with disparity map information to localise each vehicle in front of the ego-vehicle, including occluded instances, in an urban traffic environment. We also compare our approach with state-of-the-art instance segmentation methods. In the future, we will extend our work to other types of obstacles, to improve awareness and increase obstacle avoidance and path finding capabilities of a vehicle.</div><div><br></div>