Development of an embedded road boundary detection system based on deep learning

Abstract

The ability to sense the surrounding environment is an important developing technology in the field of automated vehicles. Lane line detection could determine a vehicle's travelable area. An embedded road boundary detection system based on deep learning was developed in this study. The system can detect structured and unstructured roads in a variety of situations. To obtain an image with clear lane markings, a convolution auto-encoder with the characteristics of noise reduction and reconstruction was used to remove all objects in the images except lane markings. Then, the feature points of the lane line were extracted, and the lane line was fitted with a hyperbolic model. Finally, a particle filter was used for lane tracking. The road boundary detection system was implemented on the NVIDIA Jetson TX2 platform. Three different situations, day, night, and rainy day were selected to demonstrate the performance of the proposed algorithm. Additionally, to deal with structured roads, some special scenes, such as shadows, tunnels, degenerate lane markings, and blocked lane markings, were considered. According to the experimental results, the accuracy of the proposed lane detection system for structured and unstructured roads was 90.02%.