PROJECT TITLE :

Robust Lane Detection from Continuous Driving ScenesUsing Deep Neural Networks

ABSTRACT:

For autonomous vehicles and sophisticated driver assistance systems, lane recognition in driving scenes is a critical element. Many advanced lane detection algorithms have been presented in recent years. Most approaches, on the other hand, focus on detecting the lane from a single image, which typically results in poor performance when dealing with extreme scenarios like high shadow, severe mark degradation, severe vehicle occlusion, and so on. In actuality, lanes are road structures that run in a continuous line. As a result, the lane that cannot be precisely detected in a single current frame may be inferred by combining data from past frames. To that end, we look at lane detection utilizing numerous frames from a continuous driving environment and present a hybrid deep architecture that combines the convolutional neural network (CNN) and the recurrent neural network (RNN) (RNN). A CNN block abstracts information from each frame, and the CNN features of several continuous frames with time-series properties are subsequently sent into the RNN block for feature learning and lane prediction. Extensive tests on two large-scale datasets show that the proposed method outperforms competing methods in lane detection, particularly in challenging scenarios.