Deep Learning for Road Asset Detection in Self Driving Cars

Description :

Self-driving cars or autonomous cars can sense and navigate through an environment without any driver intervention. To achieve this task, they need to be equipped with reliable sensors and an accurate algorithm to detect movable and non-movable objects around it. The cars are usually equipped with two major sensors namely, Radio Detection and Ranging (RADAR) and a Light Detection and Ranging (LIDAR). The radar is used for speed detection using Doppler effect while the LIDAR is used for position detection of static objects. In this research, we aim to use the incoming data from a LIDAR to detect static objects on the road like lanes, traffic lights, bike lanes, road edges … For that purpose, we will use deep neural networks, a method that has been yielding state of the art results in image processing. We will try to design a new network that will be able to detect and localize static road assets in real time. The detection process (pipeline) must take less than 100ms to be safely usable in an autonomous car and must have an accuracy of more than 92%.