Adaptive step detection method combining CNN-BiLSTM-SA motion pattern recognition
-
摘要: 随着位置服务(LBS)的普及，基于智能手机的行人步频检测方法对行人航迹推算(PDR)有重要影响. 针对传统步频检测方法在行人多种运动模式下计步误差大的问题，提出一种结合CNN-BiLSTM-SA运动模式识别的自适应步频检测方法. 首先根据行人行走特点划分运动模式，使用卷积神经网络(CNN)提取行人不同运动模式的局部特征，利用自注意力机制(SA)对提取的运动特征进行权重分配；再结合双向长短期记忆网络(BiLSTM)挖掘行人运动特征的前后时序关系进行分类识别；然后根据分类结果提出自适应最小峰距和自适应动态阈值两个特征约束的峰值检测算法对步频进行检测，并在步行中动态调整阈值大小. 实验结果表明：本文提出方法在8种组合运动模式下步频检测平均误差率为1.31%，与传统峰值检测相比误差率降低5.97%，同时也优于固定阈值法.
-
关键词:
- 步频检测 /
- 行人航迹推算(PDR) /
- 峰值检测 /
- 卷积神经网络(CNN) /
- 双向长短期记忆网络(BiLSTM)
Abstract: With the popularity of location based services (LBS), smartphone-based pedestrian step detection methods have important impacts on pedestrian dead reckoning (PDR). We propose an adaptive step detection method combining CNN-BiLSTM-SA motion pattern recognition to address the problem that traditional methods have large step counting errors under multiple pedestrian motion patterns. Firstly, the motion patterns are classified according to the walking characteristics of pedestrians, and the local features of different motion patterns of pedestrians are extracted by using convolutional neural network (CNN), and the weights of the extracted motion features are assigned by using self-attention (SA) mechanism, and then the bidirectional long short term memory (BiLSTM) network is combined to mine the pre-post temporal relationship of pedestrian motion features for classification and recognition. Then the peak detection algorithm with two feature constraints, adaptive minimum peak distance and adaptive dynamic threshold, is proposed to detect the step frequency according to the classification results, and the threshold size is dynamically adjusted in walking. The experimental results show that the average error rate of the proposed method for step frequency detection under eight combined motion patterns is 1.31%, which is 5.97% lower than that of the traditional peak detection, and also better than the fixed threshold method. -
表 3 5名行人步频检测结果
行人
编号实际行
走步数传统峰值检测 固定阈值TH=10 固定阈值TH=11 本文方法 正误
差数负误
差数总误
差率/%正误
差数负误
差数总误
差率/%正误
差数负误
差数总误差
率/%正误
差数负误
差数总误
差率%1 480 30 0 6.25 24 0 5.00 16 2 3.80 9 1 2.10 2 480 42 0 3.75 27 0 5.65 10 0 2.10 4 0 0.85 3 480 51 0 10.60 19 0 3.95 3 9 2.60 6 1 1.50 4 480 36 0 7.50 23 0 4.80 6 1 1.50 4 0 0.85 5 480 40 0 8.30 12 0 2.50 3 2 1.05 4 2 1.25 
下载: 导出CSV
-
[1] HE X F, JIN R C, DAI H Y. Leveraging spatial diversity for privacy-aware location-based services in mobile networks[J]. IEEE transactions on information forensics and security, 2018, 13(6): 1524-1534. DOI: 10.1109/TIFS.2018.2797023 [2] ZHAO Y C, XU J, WU J, et al. Enhancing camera-based multimodal indoor localization with device-free movement measurement using Wi-Fi[J]. IEEE internet of things journal, 2019, 7(2): 1024-1038. DOI: 10.1109/JIOT.2019.2948605 [3] HOANG M T, YUEN B, DONG X D, et al. Recurrent neural networks for accurate RSSI indoor localization[J]. IEEE internet of things journal, 2019, 6(6): 10639-10651. DOI: 10.1109/JIOT.2019.2940368 [4] 刘公绪, 史凌峰. 室内导航与定位技术发展综述[J]. 导航定位学报, 2018, 6(2): 7-14. DOI: 10.16547/j.cnki.10-1096.20180202 [5] POULOSE A, HAN D S. UWB indoor localization using deep learning LSTM networks[J]. Applied sciences, 2020, 10(18): 6290. DOI: 10.3390/app10186290 [6] 杨狄, 唐小妹, 李柏渝, 等. 基于超宽带的室内定位技术研究综述[J]. 全球定位系统, 2015, 40(5): 34-40. DOI: 10.13442/j.gnss.1008-9268.2015.05.007 [7] WU Y, ZHU H B, DU Q X, et al. A survey of the research status of pedestrian dead reckoning systems based on inertial sensors[J]. International journal of automation and computing, 2019, 16(1): 65-83. DOI: 10.1007/s11633-018-1150-y [8] QIAN J, MA J B, YING R, et al. An improved indoor localization method using smartphone inertial sensors[C]//International Conference on Indoor Positioning and Indoor Navigation, IEEE, 2013: 1-7. DOI: 10.1109/IPIN.2013.6817854 [9] TIAN Q L, SALCIC Z, KEVIN I, et al. A multi-mode dead reckoning system for pedestrian tracking using smartphones[J]. IEEE sensors journal, 2015, 16(7): 2079-2093. DOI: 10.1109/JSEN.2015.2510364 [10] YANG R, WANG B W. PACP: a position-independent activity recognition method using smartphone sensors[J]. Information, 2016, 7(4): 72. DOI: 10.3390/info7040072 [11] CHEN K X, YAO L N, ZHANG D L, et al. A semisupervised recurrent convolutional attention model for human activity recognition[J]. IEEE transactions on neural networks and learning systems, 2019, 31(5): 1747-1756. DOI: 10.1109/TNNLS.2019.2927224 [12] WANG Q, LUO H Y, XIONG H, et al. Pedestrian dead reckoning based on walking pattern recognition and online magnetic fingerprint trajectory calibration[J]. IEEE internet of things journal, 2020, 8(3): 2011-2026. DOI: 10.1109/JIOT.2020.3016146 [13] PHAM V T, NGUYEN D A, DANG N D, et al. Highly accurate step counting at various walking states using low-cost inertial measurement unit support indoor positioning system[J]. Sensors, 2018, 18(10): 3186. DOI: 10.3390/s18103186 [14] 郭丞, 吴飞, 朱海. 多场景下的行人步频自适应检测方法[J]. 全球定位系统, 2021, 46(6): 98-106. DOI: 10.12265/j.gnss.2021062101 [15] AHOLA T M. Pedometer for running activity using accelerometer sensors on the wrist[J]. Medical equipment insights, 2010(3): 1. DOI: 10.4137/MEI.S3748 [16] TUMKUR K, SUBBIAH S. Modeling human walking for step detection and stride determination by 3-axis accelerometer readings in pedometer[C]//The 4th International Conference on Computational Intelligence, Modelling and Simulation, 2012: 199-204. DOI: 10.1109/CIMSim.2012.65 [17] 陈国良, 李飞, 张言哲. 一种基于自适应波峰检测的MEMS计步算法[J]. 中国惯性技术学报, 2016, 23(3): 315-321. DOI: 10.13695/j.cnki.12-1222/o3.2015.03.007 [18] WANG B Y, LIU X L, YU B G, et al. Pedestrian dead reckoning based on motion mode recognition using a smartphone[J]. Sensors, 2018, 18(6): 1811. DOI: 10.3390/s18061811 [19] LEE S, KIM B, KIM H, et al. Inertial sensor-based indoor pedestrian localization with minimum 802.15. 4a configuration[J]. IEEE transactions on industrial informatics, 2011, 7(3): 455-466. DOI: 10.1109/TII.2011.2158832 [20] ZHANG H M, YUAN W, SHEN Q, et al. A handheld inertial pedestrian navigation system with accurate step modes and device poses recognition[J]. IEEE sensors journal, 2014, 15(3): 1421-1429. DOI: 10.1109/JSEN.2014.2363157 [21] 崔家梁, 冯朝晖, 李芹, 等. 基于CNN和RNN的像素级视频目标跟踪算法[J]. 全球定位系统, 2019, 44(3): 1-6. DOI: 10.13442/j.gnss.1008-9268.2019.03.001 [22] 陈晓雷, 孙永峰, 李策, 等. 基于卷积神经网络和双向长短期记忆的稳定抗噪声滚动轴承故障诊断[J]. 吉林大学学报(工学版), 2022, 52(2): 296-309. DOI: 10.13229/j.cnki.jdxbgxb20211031 [23] GU F, KHOSHELHAM K, VALAEE S, et al. Locomotion activity recognition using stacked denoising autoencoders[J]. IEEE internet of things journal, 2018, 5(3): 2085-2093. DOI: 10.1109/JIOT.2018.2823084 [24] WANG Q, YE L, LUO H Y, et al. Pedestrian walking distance estimation based on smartphone mode recognition[J]. Remote sensing, 2019, 11(9): 1140. DOI: 10.3390/rs11091140 [25] CHENG Y W, HU K, WU J, et al. A convolutional neural network based degradation indicator construction and health prognosis using bidirectional long short-term memory network for rolling bearings[J]. Advanced engineering informatics, 2021(48): 101247. DOI: 10.1016/J.AEI.2021.101247 [26] WANG X, CHEN G L, CAO X, et al. Robust and accurate step counting based on motion mode recognition for pedestrian indoor positioning using a smartphone[J]. IEEE sensors journal, 2021, 22(6): 4893-4907. DOI: 10.1109/JSEN.2021.3058127 [27] RUSIECKI A. Trimmed categorical cross‐entropy for deep learning with label noise[J]. Electronics letters, 2019, 55(6): 319-320. DOI: 10.1049/EL.2018.7980 -
点击查看大图
图(12) / 表(3)
计量
- 文章访问数: 200
- HTML全文浏览量: 74
- PDF下载量: 16
- 被引次数: 0
