Panoptic driving perception aims to simultaneously accomplish three perception tasks—traffic object detection, drivable area segmentation, and lane line detection—to meet the requirements of drivability prediction in autonomous driving. Existing methods usually adopt multi-task learning networks to process multiple tasks simultaneously, which have achieved certain progress in performance but generally suffer from high computational complexity as well as an imbalance between prediction accuracy and real-time performance. To address these problems, a lightweight multi-task panoramic driving perception network based on area attention, named APDP-Net, is proposed. First, the Residual-Efficient Layer Aggregation Network (R-ELAN) is introduced as the backbone network, in which the area attention module, characterized by a local and continuous attention mechanism, maintains the receptive field while reducing computational cost. Second, a GhostFPN-ELAN neck network with a feature pyramid structure is designed, introducing Ghost Convolution to reduce computation by reasonably compressing the feature space. Finally, for the lane line detection task, a Polarized Self-Attention mechanism is introduced to establish long-range connections between pixel-level features, enhancing the spatial feature representation capability of the network. Experimental results show that APDP-Net, through lightweight network design, significantly reduces computational overhead while maintaining excellent prediction performance, providing a feasible solution for edge-side autonomous driving perception.
按照惯例[16,18-20]对不同的视觉任务采取不同的评价指标。在交通目标检测中,采用召回率(Recall),以及平均精度(mean Average Precision, mAP,本文采用mAP50%,即IOU阈值为0.5时的平均精度)作为评价指标;可行驶区域分割中,采用平均交并比(mIOU)作为评价指标;车道线检测中,采用像素精度(Pixel accuracy)以及交并比(IOU)作为评价指标。在推理速度测试中,以每秒帧率(Frames Per Second,FPS)作为评价指标。以上不同方面的评价指标,其数值越大,均代表模型对应性能越优。
RENS Q, HEK M, GIRSHICKR, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI:10.1109/TPAMI.2016.2577031 .
[2]
GLEENJ, NISHIMURAK, MINEEVAT, et al. YOLOv5: You Only Look Once,Version 5[EB/OL]. [2025-05-18].
[3]
GLENNJ, CHAURASIAA, QIUJ. YOLOv8: You Only Look Once, Version 8[EB/OL]. [2023-01-10].
ZHAOH, SHIJ, QIX, et al. Pyramid Scene Parsing Network[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(2):352-363. DOI: 10.1109/TPAMI.2016.2522835 .
[6]
CHENL C, ZHUY K, PAPANDREOUG, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation[C]//Computer Vision ― ECCV 2018. Cham: Springer, 2018: 833-851. DOI:10.1007/978-3-030-01234-2_49 .
[7]
CHITTAK, PRAKASHA, JAEGERB, et al. TransFuser: Imitation with transformer-based sensor fusion for autonomous driving[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(11): 12878-12895. DOI:10.1109/TPAMI.2022.3200245 .
[8]
PANX G, SHIJ P, LUOP, et al. Spatial as deep: Spatial CNN for traffic scene understanding[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2018, 32(1): 7276-7283. DOI:10.1609/aaai.v32i1.12301 .
[9]
HOUY N, MAZ, LIUC X, et al. Learning lightweight lane detection CNNs by self attention distillation[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE Press, 2020: 1013-1021. DOI:10.1109/ICCV.2019.00110 .
[10]
QINZ Q, WANGH Y, LIX. Ultra fast structure-aware deep lane detection[C]//Computer Vision ― ECCV 2020. Cham: Springer, 2020: 276-291. DOI:10.1007/978-3-030-58586-0_17 .
[11]
TABELINIL, BERRIELR, PAIXÃOT M, et al. Keep your eyes on the lane: Real-time attention-guided lane detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2021: 294-302. DOI:10.1109/CVPR46437.2021.00036 .
[12]
LIUL Z, CHENX H, ZHUS Y, et al. CondLaneNet: A top-to-down lane detection framework based on conditional convolution[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE Press, 2022: 3753-3762. DOI:10.1109/ICCV48922.2021.00375 .
[13]
ZHENGT, HUANGY F, LIUY, et al. CLRNet: Cross layer refinement network for lane detection[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2022: 888-897. DOI:10.1109/CVPR52688.2022.00097 .
[14]
TEICHMANNM, WEBERM, ZÖLLNERM, et al. MultiNet: Real-time joint semantic reasoning for autonomous driving[C]//2018 IEEE Intelligent Vehicles Symposium (Ⅳ). New York: IEEE Press, 2018: 1013-1020. DOI:10.1109/IVS.2018.8500504 .
[15]
QIANY Q, DOLANJ M, YANGM. DLT-net: Joint detection of drivable areas, lane lines, and traffic objects[J]. IEEE Transactions on Intelligent Transportation Systems, 2020, 21(11): 4670-4679. DOI:10.1109/TITS.2019.2943777 .
[16]
WUD, LIAOM W, ZHANGW T, et al. YOLOP: You only look once for panoptic driving perception[J]. Machine Intelligence Research, 2022, 19(6): 550-562. DOI:10.1007/s11633-022-1339-y .
WANGJ Y, JONATHAN WUQ M, ZHANGN. You only look at once for real-time and generic multi-task[EB/OL]. 2023: arXiv: 2310.01641. DOI: 10.1109/tvt.2024.3394350 .
[21]
HUJ, SHENL, ALBANIES, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023. DOI:10.1109/TPAMI.2019.2913372 .
HANK, WANGY H, TIANQ, et al. GhostNet: More features from cheap operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2020: 1577-1586. DOI:10.1109/CVPR42600.2020.00165 .
[24]
LIUH J, LIUF Q, FANX Y, et al. Polarized self-attention: Towards high-quality pixel-wise regression[EB/OL]. 2021: arXiv: 2107.00782. DOI: 10.1016/j.neucom.2022.07.054 .
[25]
LIUZ, LINY T, CAOY, et al. Swin transformer: Hierarchical vision transformer using shifted windows[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE Press, 2022: 9992-10002. DOI:10.1109/ICCV48922.2021.00986 .
[26]
DONGX Y, BAOJ M, CHEND D, et al. CSWin transformer: A general vision transformer backbone with cross-shaped windows[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2022: 12114-12124. DOI:10.1109/CVPR52688.2022.01181 . [LinkOut]
[27]
PASZKEA, CHAURASIAA, KIMS, et al. ENet: A deep neural network architecture for real-time semantic segmentation[EB/OL]. 2016: arXiv: 1606.02147. DOI: 10.48550/arXiv.1606.02147 .