Lane detection is a feature of autonomous vehicles and is an essential technology for environmental awareness. While many academics have successfully used Deep Learning (DL) for lane detection, their application cases were rather straightforward. When a lane is blocked, lost, or has other problems, the accuracy of lane recognition diminishes significantly. In this paper, to solve the above-mentioned problems, a Hyperbolic Neural Network (HNN) with a Bi-directional Long Short-Term Memory (B-LSTM) model is proposed for efficient autonomous vehicle lane detection. At first, hyperbolic space is used to identify the space changes in the frame, and the HNN is used to extract the spatial information in the frames. Then, finally, the B-LSTM is used to perform the sequential data to detect the autonomous vehicle lane in an effective way. The implemented HNN with the B-LSTM model demonstrates superior performance, achieving a Precision of 0.0311, Recall of 0.0253, F1-Score of 97.57%, and Accuracy of 96.74%. When compared to previous models like the Lightweight Attention Deep Neural Network (LA-DNN), ResNest and Self-attention Distillation (RS-Lane), and Convolutional Neural Network-Multi Layer Perceptron (CNN-MLP).
Autonomous Vehicle Lane Detection Using Hyperbolic Neural Network with Bi-Directional Long Short-Term Memory
04.12.2023
330733 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Taylor & Francis Verlag | 2023
|