基于特征通道和空間聯合注意機制的遮擋行人檢測方法
doi: 10.11999/JEIT190606 cstr: 32379.14.JEIT190606
-
1.
重慶郵電大學工業(yè)物聯網與網絡化控制教育部重點實驗室 重慶 400065
-
2.
重慶郵電大學通信與信息工程學院 重慶 400065
Occluded Pedestrian Detection Based on Joint Attention Mechanism of Channel-wise and Spatial Information
-
1.
Key Laboratory of Industrial Internet of Things & Network Control, Ministry of Education, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
-
2.
School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China
-
摘要:
遮擋是行人檢測任務中導致漏檢發(fā)生的主要原因之一，對檢測器性能造成了不利影響。為了增強檢測器對于遮擋行人目標的檢測能力，該文提出一種基于特征引導注意機制的單級行人檢測方法。首先，設計一種特征引導注意模塊，在保持特征通道間的關聯性的同時保留了特征圖的空間信息，引導模型關注遮擋目標可視區(qū)域；然后，通過注意模塊融合淺層和深層特征，從而提取到行人的高層語義特征；最后，將行人檢測作為一種高層語義特征檢測問題，通過激活圖的形式預測得到行人位置和尺度，并生成最終的預測邊界框，避免了基于先驗框的預測方式所帶來的額外參數設置。所提方法在CityPersons數據集上進行了測試，并在Caltech數據集上進行了跨數據集實驗。結果表明該方法對于遮擋目標檢測準確度優(yōu)于其他對比算法。同時該方法實現了較快的檢測速度，取得了檢測準確度和速度的平衡。
Abstract:Pedestrian detector performance is damaged because occlusion often leads to missed detection. In order to improve the detector's ability to detect pedestrian, a single-stage detector based on feature-guided attention mechanism is proposed. Firstly, a feature attention module is designed, which preserves the association between the feature channels while retaining spatial information, and guides the model to focus on visible region. Secondly, the attention module is used to fuse shallow and deep features, then high-level semantic features of pedestrians are extracted. Finally, pedestrian detection is treated as a high-level semantic feature detection problem. Pedestrian location and scale are obtained through heat map prediction, then the final prediction bounding box is generated. This way, the proposed method avoids the extra parameter settings of the traditional anchor-based method. Experiments show that the proposed method is superior to other comparison algorithms for the accuracy of occlusion target detection on CityPersons and Caltech pedestrian database. At the same time, the proposed method achieves a faster detection speed and a better balance between detection accuracy and speed.
-
表 1 驗證實驗條件設置
R (Reasonable) HO (Heavy Occlusion) R+HO (Reasonable+Heavy Occlusion) $v \in [0.65,\infty )$ $v \in [0.20,0.65]$ $v \in [0.20,\infty )$ 
下載: 導出CSV
表 2 注意網絡驗證結果MR–2(%)
方法 R HO R+HO 文獻[16] 16.0 56.7 38.2 Baseline 12.1 41.1 38.1 Baseline+CA 11.8 39.2 37.8 Baseline+CA+SA 11.6 38.5 37.3
下載: 導出CSV
表 3 CityPersons數據集測試結果MR-2(%)
方法 主干網絡 Reasonable Heavy Partial Bare 測試時間(s) OR-CNN[11] VGG-16 12.8 55.7 15.3 6.7 – FasterRCNN[21] VGG-16 15.4 – – – – ALFNet[8] ResNet-50 12.0 51.9 11.4 8.4 0.27 CSP[9] ResNet-50 11.0 49.3 10.4 7.3 0.33 CAFL[13] ResNet-50 11.4 50.4 12.1 7.6 – PedJointNet[14] ResNet-50 13.5 52.1 – – – TLL[20] ResNet-50 15.5 53.6 17.2 10.0 – RepLoss[23] ResNet-50 13.2 56.9 16.8 7.6 – 本文方法 ResNet-50 11.6 47.6 9.8 7.5 0.22
下載: 導出CSV
-
張功國, 吳建, 易億, 等. 基于集成卷積神經網絡的交通標志識別[J]. 重慶郵電大學學報: 自然科學版, 2019, 31(4): 571–577. doi: 10.3979/j.issn.1673-825X.2019.04.019ZHANG Gongguo, WU Jian, YI Yi, et al. Traffic sign recognition based on ensemble convolutional neural network[J]. Journal of Chongqing University of Posts and Telecommunications:Natural Science Edition, 2019, 31(4): 571–577. doi: 10.3979/j.issn.1673-825X.2019.04.019 種衍文, 匡湖林, 李清泉. 一種基于多特征和機器學習的分級行人檢測方法[J]. 自動化學報, 2012, 38(3): 375–381. doi: 10.3724/SP.J.1004.2012.00375CHONG Yanwen, KUANG Hulin, and LI Qingquan. Two-stage pedestrian detection based on multiple features and machine learning[J]. Acta Automatica Sinica, 2012, 38(3): 375–381. doi: 10.3724/SP.J.1004.2012.00375 劉威, 段成偉, 遇冰, 等. 基于后驗HOG特征的多姿態(tài)行人檢測[J]. 電子學報, 2015, 43(2): 217–224. doi: 10.3969/j.issn.0372-2112.2015.02.002LIU Wei, DUAN Chengwei, YU Bing, et al. Multi-pose pedestrian detection based on posterior HOG feature[J]. Acta Electronica Sinica, 2015, 43(2): 217–224. doi: 10.3969/j.issn.0372-2112.2015.02.002 REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[C]. 2015 Advances in Neural Information Processing Systems, Montreal, Canada, 2015: 91–99. LI Jianan, LIANG Xiaodan, SHEN Shengmei, et al. Scale-aware fast R-CNN for pedestrian detection[J]. IEEE Transactions on Multimedia, 2018, 20(4): 985–996. doi: 10.1109/TMM.2017.2759508 王進, 陳知良, 李航, 等. 一種基于增量式超網絡的多標簽分類方法[J]. 重慶郵電大學學報: 自然科學版, 2019, 31(4): 538–549. doi: 10.3979/j.issn.1673-825X.2019.04.015WANG Jin, CHEN Zhiliang, LI Hang, et al. Hierarchical multi-label classification using incremental hypernetwork[J]. Journal of Chongqing University of Posts and Telecommunications:Natural Science Edition, 2019, 31(4): 538–549. doi: 10.3979/j.issn.1673-825X.2019.04.015 GIRSHICK R. Fast R-CNN[C]. 2015 IEEE International Conference on Computer Vision, Santiago, Chile, 2015: 1440–1448. LIU Wei, LIAO Shengcai, HU Weidong, et al. Learning efficient single-stage pedestrian detectors by asymptotic localization fitting[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 618–634. LIU Wei, LIAO Shengcai, REN Weiqiang, et al. High-level semantic feature detection: A new perspective for pedestrian detection[C]. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 5182–5191. ZHANG Shanshan, BENENSON R, OMRAN M, et al. How far are we from solving pedestrian detection?[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 1259–1267. ZHANG Shifeng, WEN Longyin, BIAN Xiao, et al. Occlusion-aware R-CNN: detecting pedestrians in a crowd[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 637–653. OUYANG Wanli, ZHOU Hui, LI Hongsheng, et al. Jointly learning deep features, deformable parts, occlusion and classification for pedestrian detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(8): 1874–1887. doi: 10.1109/TPAMI.2017.2738645 FEI Chi, LIU Bin, CHEN Zhu, et al. Learning pixel-level and instance-level context-aware features for pedestrian detection in crowds[J]. IEEE Access, 2019, 7: 94944–94953. doi: 10.1109/ACCESS.2019.2928879 LIN C Y, XIE Hongxia, and ZHENG Hua. PedJointNet: Joint head-shoulder and full body deep network for pedestrian detection[J]. IEEE Access, 2019, 7: 47687–47697. doi: 10.1109/ACCESS.2019.2910201 ZHANG Shanshan, YANG Jian, and SCHIELE B. Occluded pedestrian detection through guided attention in CNNs[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 6995–7003. ZHU Chenchen, HE Yihui, and SAVVIDES M. Feature selective anchor-free module for single-shot object detection[C]. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 840–849. LIN T Y, DOLLáR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 936–944. CHEN Long, ZHANG Hanwang, XIAO Jun, et al. SCA-CNN: Spatial and channel-wise attention in convolutional networks for image captioning[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 6298–6306. WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional block attention module[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 3–19. SONG Tao, SUN Leiyu, XIE Di, et al. Small-scale pedestrian detection based on topological line localization and temporal feature aggregation[C]. The 15th European Conference on Computer Vision, Munich, Germany, 2018: 536–551. ZHANG Shanshan, BENENSON R, and SCHIELE B. Citypersons: A diverse dataset for pedestrian detection[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 4457–4465. DOLLAR P, WOJEK C, SCHIELE B, et al. Pedestrian detection: An evaluation of the state of the art[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(4): 743–761. doi: 10.1109/TPAMI.2011.155 WANG Xinlong, XIAO Tete, JIANG Yuning, et al. Repulsion loss: Detecting pedestrians in a crowd[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 7774–7783. -
計量
- 文章訪問數: 3618
- HTML全文瀏覽量: 1465
- PDF下載量: 166
- 被引次數: 0
下載:
