Combining Deep Learning with traditional algorithms in autonomous cars

Abstract

Research of autonomous technologies in modern vehicles are being conducted as never before. For a long time, traditional computer vision based algorithms has been the primary method for analyzing camera footage, used for assisting safety functions, where decision making have been a product of manually constructed behaviours. During the last few years deep learning has demonstrated its extraordinary capabilities for both visual recognition and decision making in end-to-end systems. In this report we propose a solution of introducing redundancy by combining deep learning methods with traditional computer vision based techniques for minimizing unsafe behaviour in autonomous vehicles. The system consists of a computer vision based lane detection algorithm in combination with a fully connected Deep Neural Network, and combines the advantages of both technologies by constructing a control algorithm responsible for consolidating the sub systems calculations of the correct steering angle, used to keep the vehicle within the lane markings of the road. The solution proposed show that we can increase the performance of our system by applying a combination of the two technologies in a simulator resulting in a safer system than we could achieve with the technologies separately.