UrbanFM: Inferring Fine-Grained Urban Flows

KDD, 2019.

Cited by: 24|Bibtex|Views103
EI
Other Links: dblp.uni-trier.de|academic.microsoft.com|dl.acm.org|arxiv.org
Weibo:
Urban Flow Magnifier has addressed the two challenges that are specific to this problem, i.e., the spatial correlation as well as the complexities of external factors, by leveraging the original distributional upsampling module and the external factor fusion subnet

Abstract:

Urban flow monitoring systems play important roles in smart city efforts around the world. However, the ubiquitous deployment of monitoring devices, such as CCTVs, induces a long-lasting and enormous cost for maintenance and operation. This suggests the need for a technology that can reduce the number of deployed devices, while preventing...More

Code:

Data:

0
Introduction
  • The fine-grained urban flow monitoring system is a crucial component of the information infrastructure system in smart cities, which lays the foundation for urban planning and various applications such as traffic management.
  • With the rapid development of smart cities on a worldwide scale, the cost of manpower and energy will become a prohibitive factor for the further intelligentization of the Earth.
  • To reduce such expense, people require a novel technology which allows cutting the number of deployed sensors while, most importantly, keeping the original data granularity unchanged.
  • How to approximate the original fine-grained information from available coarse-grained data becomes an urgent problem
Highlights
  • The fine-grained urban flow monitoring system is a crucial component of the information infrastructure system in smart cities, which lays the foundation for urban planning and various applications such as traffic management
  • The advances of Urban Flow Magnifier (UrbanFM)-ne over all baselines indicate that the distribution upsampling in our inference network plays a leading role in improving the inference performance; the advances of UrbanFM over UrbanFM-ne support that the combination with external subnet enhances the model by incorporating external factors
  • (2) Image super-resolution methods outdo the heuristic method Historical Average (HA) on Root Mean Square Error (RMSE) while show deteriorated scores on Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). This can be attributed to two reasons: first, neural network methods are dedicated to performing well on RMSE as it is the training objective; second, HA preserves the spatial correlation for fine-grained flow maps while the others fail to do so
  • A piece of further evidence can be seen from the comparison between UrbanFM-ne and SRResNet, where the former model has a similar structure as SRResNet except for the distributional upsampling module, which makes it surpass its counterpart
  • UrbanFM has addressed the two challenges that are specific to this problem, i.e., the spatial correlation as well as the complexities of external factors, by leveraging the original distributional upsampling module and the external factor fusion subnet
  • Various empirical studies and visualizations have confirmed the advantages of UrbanFM on both efficiency and effectiveness
Methods
  • MEAN HA

    SRCNN ESPCN DeepSD VDSR SRResNet UrbanFM-ne UrbanFM

    VDSR [11]: Since both SRCNN and ESPCN follow a three-stage architecture, they have several drawbacks such as slow convergence speed and limited representation ability.
Results
  • Results on TaxiBJ

    Model Comparison In this subsection, the authors compare the model effectiveness against the baselines.
  • This can be attributed to two reasons: first, neural network methods are dedicated to performing well on RMSE as it is the training objective; second, HA preserves the spatial correlation for fine-grained flow maps while the others fail to do so.
  • One important trait of the HappyValley dataset is that it contains more spikes on the fine-grained flow distribution, which results in a much larger RMSE score versus that in the TaxiBJ task.
  • This proves that UrbanFM works on the large-scale scenario, but is adaptable to smaller areas, which concludes the empirical studies
Conclusion
  • The authors have formalized the fine-grained urban flow inference problem and presented a deep neural network-based method (UrbanFM) to solve it.
  • UrbanFM has addressed the two challenges that are specific to this problem, i.e., the spatial correlation as well as the complexities of external factors, by leveraging the original distributional upsampling module and the external factor fusion subnet.
  • Various empirical studies and visualizations have confirmed the advantages of UrbanFM on both efficiency and effectiveness.
  • The authors will explore more on improving the model structure, and pay more attention to reducing errors in hard regions
Summary
  • Introduction:

    The fine-grained urban flow monitoring system is a crucial component of the information infrastructure system in smart cities, which lays the foundation for urban planning and various applications such as traffic management.
  • With the rapid development of smart cities on a worldwide scale, the cost of manpower and energy will become a prohibitive factor for the further intelligentization of the Earth.
  • To reduce such expense, people require a novel technology which allows cutting the number of deployed sensors while, most importantly, keeping the original data granularity unchanged.
  • How to approximate the original fine-grained information from available coarse-grained data becomes an urgent problem
  • Objectives:

    The authors aim to infer the real-time and finegrained crowd flows throughout a city based on coarse-grained observations.
  • The authors' goal is to infer the real-time and spatially fine-grained flows from observed coarsegrained data on a citywide scale with many other regions.
  • Apart from the above applications, the authors aim to solve a novel problem (FUFI) on urban flows in this study
  • Methods:

    MEAN HA

    SRCNN ESPCN DeepSD VDSR SRResNet UrbanFM-ne UrbanFM

    VDSR [11]: Since both SRCNN and ESPCN follow a three-stage architecture, they have several drawbacks such as slow convergence speed and limited representation ability.
  • Results:

    Results on TaxiBJ

    Model Comparison In this subsection, the authors compare the model effectiveness against the baselines.
  • This can be attributed to two reasons: first, neural network methods are dedicated to performing well on RMSE as it is the training objective; second, HA preserves the spatial correlation for fine-grained flow maps while the others fail to do so.
  • One important trait of the HappyValley dataset is that it contains more spikes on the fine-grained flow distribution, which results in a much larger RMSE score versus that in the TaxiBJ task.
  • This proves that UrbanFM works on the large-scale scenario, but is adaptable to smaller areas, which concludes the empirical studies
  • Conclusion:

    The authors have formalized the fine-grained urban flow inference problem and presented a deep neural network-based method (UrbanFM) to solve it.
  • UrbanFM has addressed the two challenges that are specific to this problem, i.e., the spatial correlation as well as the complexities of external factors, by leveraging the original distributional upsampling module and the external factor fusion subnet.
  • Various empirical studies and visualizations have confirmed the advantages of UrbanFM on both efficiency and effectiveness.
  • The authors will explore more on improving the model structure, and pay more attention to reducing errors in hard regions
Tables
  • Table1: Dataset Description
  • Table2: Results comparisons on TaxiBJ over different time spans (P1-P4)
  • Table3: Results for different M-F settings
  • Table4: Results comparison on Happy Valley
  • Table5: The details of partition over two datasets
  • Table6: Details settings of Inference Network in Figure 3, where settings k-s-n means the size of kernel, stride and number of filters in a certain convolutional layer. We omit the batch size in the format of output for simplicity
  • Table7: Embedding setting of external factors
  • Table8: Details of External Factor Fusion in Figure 3
Download tables as Excel
Related work
  • 5.1 Image Super-Resolution

    Single image super-resolution (SISR), which aims to recover a highresolution (HR) image from a single low-resolution (LR) image, has gained increasing research attention for decades. This task finds direct applications in many areas such as face recognition [5], fine-grained crowdsourcing [24] and HDTV [18]. Over years, many SISR algorithms have been developed in the computer vision community. To tackle the SR problem, early techniques focused on interpolation methods such as bicubic interpolation and Lanczos resampling [3]. Also, several studies utilized statistical image priors (a) Studied Area

    Office Area Residence Restaurant (b) 10:00 weekday (c) 21:00 weekday (d) 10:00 weekend [22, 23] to achieve better performances. Advanced works aimed at learning the non-linear mapping between LR and HR images with neighbor embedding [1] and sparse coding [25, 28]. However, these approaches are still inadequate to reconstruct realistic and fine-grained textures of images.
Funding
  • This work was supported by the National Natural Science Foundation of China Grant No 61672399, No U1609217 and No 61773324, as well as A*STAR SERC PSF under grant 152120008
Reference
  • Hong Chang, Dit-Yan Yeung, and Yimin Xiong. 2004. Super-resolution through neighbor embedding. In CVPR.
    Google ScholarFindings
  • Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. 2016. Image superresolution using deep convolutional networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 38, 2 (2016), 295–307.
    Google ScholarLocate open access versionFindings
  • Claude E Duchon. 1979. Lanczos filtering in one and two dimensions. Journal of Applied Meteorology 18, 8 (1979), 1016–1022.
    Google ScholarLocate open access versionFindings
  • Zipei Fan, Xuan Song, Ryosuke Shibasaki, and Ryutaro Adachi. 2015. CityMomentum: an online approach for crowd behavior prediction at a citywide level. In Proceedings of the ACM International Joint Conference on Pervasive and Ubiquitous Computing. 559–569.
    Google ScholarLocate open access versionFindings
  • Bahadir K Gunturk, Aziz Umit Batur, Yucel Altunbasak, Monson H Hayes, and Russell M Mersereau. 2003. Eigenface-domain super-resolution for face recognition. IEEE Transactions on Image Processing 12, 5 (2003), 597–606.
    Google ScholarLocate open access versionFindings
  • Richard HR Hahnloser, Rahul Sarpeshkar, Misha A Mahowald, Rodney J Douglas, and H Sebastian Seung. 2000. Digital selection and analogue amplification coexist in a cortex-inspired silicon circuit. Nature 405, 6789 (2000), 947.
    Google ScholarLocate open access versionFindings
  • Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. In CVPR. 770–778.
    Google ScholarLocate open access versionFindings
  • Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Identity mappings in deep residual networks. In ECCV. 630–645.
    Google ScholarLocate open access versionFindings
  • Minh X Hoang, Yu Zheng, and Ambuj K Singh. 2016. FCCF: forecasting citywide crowd flows based on big data. In Proceedings of the ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. 6.
    Google ScholarLocate open access versionFindings
  • Sergey Ioffe and Christian Szegedy. 2015. Batch normalization: Accelerating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167 (2015).
    Findings
  • Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. 2016. Accurate image superresolution using very deep convolutional networks. In CVPR. 1646–1654.
    Google ScholarFindings
  • Diederik P Kingma and Jimmy Ba. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014).
    Findings
  • Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. 2015. Deep learning. Nature 521, 7553 (2015), 436.
    Google ScholarLocate open access versionFindings
  • Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew P Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. 2017. Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network.. In CVPR, Vol. 2. 4.
    Google ScholarLocate open access versionFindings
  • Sang Keon Lee, Heeseo Rain Kwon, HeeAh Cho, Jongbok Kim, and Donju Lee. 2016. International Case Studies of Smart Cities: Anyang, Republic of Korea. (2016).
    Google ScholarLocate open access versionFindings
  • Yuxuan Liang, Songyu Ke, Junbo Zhang, Xiuwen Yi, and Yu Zheng. 2018. GeoMAN: Multi-level Attention Networks for Geo-sensory Time Series Prediction. In Proceedings of the International Joint Conference on Artificial Intelligence.
    Google ScholarLocate open access versionFindings
  • Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee. 20Enhanced deep residual networks for single image super-resolution. In The IEEE Conference on Computer Vision and Pattern Recognition Workshops, Vol. 1. 4.
    Google ScholarLocate open access versionFindings
  • Sung Cheol Park, Min Kyu Park, and Moon Gi Kang. 2003. Super-resolution image reconstruction: a technical overview. IEEE Signal Processing Magazine 20, 3 (2003), 21–36.
    Google ScholarLocate open access versionFindings
  • Wenzhe Shi, Jose Caballero, Ferenc Huszár, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. 2016. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In ICCV. 1874–1883.
    Google ScholarFindings
  • Karen Simonyan and Andrew Zisserman. 2014. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014).
    Findings
  • Xuan Song, Quanshi Zhang, Yoshihide Sekimoto, and Ryosuke Shibasaki. 2014. Prediction of human emergency behavior and their mobility following largescale disaster. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 5–14.
    Google ScholarLocate open access versionFindings
  • Jian Sun, Zongben Xu, and Heung-Yeung Shum. 2008. Image super-resolution using gradient profile prior. In CVPR. 1–8.
    Google ScholarLocate open access versionFindings
  • Yu-Wing Tai, Shuaicheng Liu, Michael S Brown, and Stephen Lin. 2010. Super resolution using edge prior and single image detail synthesis. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2400–2407.
    Google ScholarLocate open access versionFindings
  • Matt W Thornton, Peter M Atkinson, and DA Holland. 2006. Sub-pixel mapping of rural land cover objects from fine spatial resolution satellite sensor imagery using super-resolution pixel-swapping. International Journal of Remote Sensing 27, 3 (2006), 473–491.
    Google ScholarLocate open access versionFindings
  • Radu Timofte, Vincent De Smet, and Luc Van Gool. 2014. A+: Adjusted anchored neighborhood regression for fast super-resolution. In Asian Conference on Computer Vision. 111–126.
    Google ScholarLocate open access versionFindings
  • Thomas Vandal, Evan Kodra, Sangram Ganguly, Andrew Michaelis, Ramakrishna Nemani, and Auroop R Ganguly. 2017. Deepsd: Generating high resolution climate change projections through single image super-resolution. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 1663–1672.
    Google ScholarLocate open access versionFindings
  • Leye Wang, Xu Geng, Xiaojuan Ma, Feng Liu, and Qiang Yang. 2018. Crowd Flow Prediction by Deep Spatio-Temporal Transfer Learning. arXiv preprint arXiv:1802.00386 (2018).
    Findings
  • Jianchao Yang, John Wright, Thomas S Huang, and Yi Ma. 2010. Image superresolution via sparse representation. IEEE Transactions on Image Processing 19, 11 (2010), 2861–2873.
    Google ScholarLocate open access versionFindings
  • Junbo Zhang, Yu Zheng, and Dekang Qi. 2017. Deep Spatio-Temporal Residual Networks for Citywide Crowd Flows Prediction. In The AAAI Conference on Artificial Intelligence. 1655–1661.
    Google ScholarLocate open access versionFindings
  • Junbo Zhang, Yu Zheng, Dekang Qi, Ruiyuan Li, and Xiuwen Yi. 2016. DNN-based prediction model for spatio-temporal data. In Proceedings of the ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. 92.
    Google ScholarLocate open access versionFindings
  • Yu Zheng, Licia Capra, Ouri Wolfson, and Hai Yang. 2014. Urban computing: concepts, methodologies, and applications. ACM Transactions on Intelligent Systems and Technology 5, 3 (2014), 38.
    Google ScholarLocate open access versionFindings
Full Text
Your rating :
0

 

Tags
Comments