Table 1 Comparisons among different methods of lane mark line detection.
From: Efficient spatial and channel net for lane marker detection based on self-attention and row anchor
Methods | Strength | Limitation | Performance |
|---|---|---|---|
Tradition approaches based on vision | (1) Simple and convenient | (1) Only depend on edges, color, thickness and shape to detect lanes | (1) For simple scenarios |
(2) The computing power of the embedded platform is not relatively high | (2) The adaptability of the algorithm is not strong | (2) Accuracy and F1 indicators are not very high | |
| Â | (3) Result in a lot of work and low robustness. When the driving environment changes significantly, the effect of lane line detection is not good | (3) Speed of FPS is generally is generally fast | |
(4) Be sensitive to changes in light, weather conditions and noise. When the external environment changes significantly, many traditional lane detection systems fail | Â | ||
Segmentation approach using CNN | (1) Lightweight and efficient networks | (1) Require more computing resources | (1) Higher accuracy and better robustness |
(2) Particularly suitable for embedded systems and real-time applications | (2) The model is large and the processing speed is generally slow | (2) Suitable for urban roads | |
| Â | (3) Under strong obstacles, performance was poor and prior knowledge of the lane line was not fully utilized | (3) Speed of FPS is not too high | |
With CNN + attention | (1) Emphasizing local information and global information | (1) Higher computing power requirements | (1) Suitable for complex scenarios |
(2) Easily to handle the obstruction, lack or weak display of the lane lines in complex scenes | (2) Network model is more complicated and big | (2) Higher accuracy and F1 indicators | |
(3) Better real -time | (3) Requiring more memory resources | (3) Speed of FPS is high in the case where the computing power is guaranteed |
