Lane Detection

Lane detection is a problem where computer vision algorithms must identify and track the boundaries of driving lanes in images or video in real-time, even under challenging conditions such as poor lighting, shadows, glare, complex lane markings, and varying road shapes. This information is crucial for self-driving cars and advanced driver assistance systems (ADAS) as it allows them to understand the road layout, determine their position within the lane, and guide the vehicle safely.

Previous work in (Widianto, 2020) utilizes Hough transform for doing the Lane Detection. It produces good result in a normal and straight road condition. However, in some challenging scenarios such as in rain or nighttime, the performance was degrading. Therefore, in this study the hybrid CNN-RNN method was used to solve the lane detection problem. This combination of CNN-RNN aims to take advantage of the feature extraction capabilities of CNN and temporal information processing of RNN. Two kinds of architectures are used, namely SegNet-ConvLSTM and UNet-ConvLSTM. The performance of these two architectures was examined for three types of weather conditions and three types of road conditions. The performance metric used in this study is the F1-score. The average F1-score of SegNet-ConvLSTM and UNet-ConvLSTM are 0.6751 and 0.6742, respectively. This is four times higher when it is compared with Hough transform which only managed to get 0.1545 in its F1 score.

The following are the results and comparison of some different algorithms.