Chinese Journal of Liquid Crystals and Displays, Volume. 39, Issue 4, 543(2024)
Lane detection algorithm based on ARM embedded platform
Fig. 8. Actual application scenarios of Nvidia SUB development board
Fig. 10. Lane line detection results for heavily shaded road scene in ARM platform
Fig. 11. Lane line detection results for rainy day scenario in ARM platform
Fig. 12. Lane line detection results for curved road scene in the ARM platform
Fig. 13. Lane line detection results for the scene of vehicle in front occlusion in ARM platform
Fig. 14. Lane line detection results for the scene with road signs on the ground
Fig. 15. Lane line detection results for the scene of ground contamination
Table 1. U-Segnet network coding structure
View tableView in ArticleTable 1. U-Segnet network coding structure
类型 滤波器 卷积核大小 输出 输入128×256×3 卷积 64 3×3 128×256 卷积 64 3×3 128×256 池化 卷积 128 3×3 64×128 卷积 128 3×3 64×128 池化 卷积 256 3×3 32×64 卷积 256 3×3 32×64 卷积 256 3×3 32×64
Table 2. Structure of Tusimple# data set
View tableView in ArticleTable 2. Structure of Tusimple# data set
数据集 包含 被标记帧序号 被标记总帧数 Train Tusimple 13th and 20th 7 252 Ours 13th and 20th 2 296 Test Testset #1 13th and 20th 540 Testset #2 all frames 728
Table 3. Ablation results
View tableView in ArticleTable 3. Ablation results
实验 U-Segnet算法模块 SE模块 分数 实验1 - - 0.72 实验2 √ - 0.80 实验3 - √ 0.73 实验4 √ √ 0.85
Table 4. Comparison of detection accuracy of Tusimple extended data set
View tableView in ArticleTable 4. Comparison of detection accuracy of Tusimple extended data set
方法 准确率/% 时延/s Segnet［ 10 ］92.38 0.050 Unet［ 11 ］94.34 0.040 Segnet-ConvLSTM［ 13 ］95.29 0.125 Unet-ConvLSTM［ 13 ］95.67 0.067 ESNet［ 16 ］90.08 0.018 Fast-SCNN［ 17 ］89.34 0.022 FPENet［ 18 ］87.65 0.016 Ours 98.03 0.017
Table 5. Comparison of reasoning results of different models
View tableView in ArticleTable 5. Comparison of reasoning results of different models
推理模型 语言 单帧检测时间/s FPS Pytorch模型 Python 0.083 9 Trt模型 python 0.020 50
Table 6. Lane detection results in different traffic scene
View tableView in ArticleTable 6. Lane detection results in different traffic scene
场景类别 选取图像帧序号 车道线总帧数 正确检测帧数 准确率/% FPS 1 178 308 301 97.64 51 2 256 466 456 97.89 53 3 254 424 398 93.87 54 4 364 597 565 94.59 54 5 223 494 456 92.36 50 6 587 896 541 90.41 49 7 355 456 398 87.26 45 8 260 598 532 89.03 47
Get Citation
Copy Citation Text
Tiantian GUAN, Fan YANG. Lane detection algorithm based on ARM embedded platform[J]. Chinese Journal of Liquid Crystals and Displays, 2024, 39(4): 543
Category: Research Articles
Received: Apr. 13, 2023
Accepted: --
Published Online: May. 28, 2024
The Author Email: Fan YANG (15620831298@qq.com)
