Litedge: Towards Light-weight Edge Computing for Efficient Wireless Surveillance System

Wireless surveillance systems are rapidly gaining popularity due to their easier deployability and improved performance. However, cameras inside are generating a large amount of data, which brings challenges to the transmission through resource-constrained wireless networks. Observing that most collected consecutive frames are redundant with few objects of interest (OoIs), the filtering of these frames can dramatically relieve the transmission pressure. Additionally, real-world environment may bring shielding or blind areas in videos, which notoriously affects the accuracy of frame analysis. The collaboration between cameras facing at different angles can compensate for such accuracy loss. In this work, we present Litedge, a light-weight edge computing strategy to improve the QoS (i. e., the latency and accuracy) of wireless surveillance systems. Two main modules are designed on edge cameras: (i) the light-weight video compression module for frame filtering, mainly realized by model compression and convolutional acceleration; and (ii) the collaborative validation module for error compensation between the master-slave camera pair. We also implement an enhanced surveillance system prototype from real-time monitoring and pre-processing on edge cameras to the backend data analysis on a server. Experiments based on real-world collected videos show the efficiency of Litedge. It achieves 82% transmission latency reduction with a maximal 0.119s additional processing delay, compared with the full video transmission. Remarkably, 91.28% of redundant frames are successfully filtered out, greatly reducing the transmission burden. Litedge outperforms state-of-the-art light-weight AI models and video compression methods by balancing the QoS balance ratio between accuracy and latency.

[1]  Xing Zhang,et al.  A Survey on Mobile Edge Networks: Convergence of Computing, Caching and Communications , 2017, IEEE Access.

[2]  Guangjie Han,et al.  Edge Computing-Based Intelligent Manhole Cover Management System for Smart Cities , 2018, IEEE Internet of Things Journal.

[3]  Haohong Wang,et al.  Video Surveillance Over Wireless Sensor and Actuator Networks Using Active Cameras , 2011, IEEE Transactions on Automatic Control.

[4]  Yacine Challal,et al.  Energy efficiency in wireless sensor networks: A top-down survey , 2014, Comput. Networks.

[5]  Yann LeCun,et al.  Fast Training of Convolutional Networks through FFTs , 2013, ICLR.

[6]  Xiangyu Zhang,et al.  Channel Pruning for Accelerating Very Deep Neural Networks , 2017, 2017 IEEE International Conference on Computer Vision (ICCV).

[7]  Debin Zhao,et al.  Up-sampling Dependent Frame Rate Reduction for Low Bit-Rate Video Coding , 2011, 2011 Data Compression Conference.

[8]  Gian Luca Foresti,et al.  An integrated surveillance system for outdoor security , 2005, IEEE Conference on Advanced Video and Signal Based Surveillance, 2005..

[9]  Bo Chen,et al.  MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications , 2017, ArXiv.

[10]  Danna Zhou,et al.  d. , 1934, Microbial pathogenesis.

[11]  Mohan S. Kankanhalli,et al.  Multi-Camera Coordination and Control in Surveillance Systems , 2015, ACM Trans. Multim. Comput. Commun. Appl..

[12]  Joan L. Mitchell,et al.  MPEG Video: Compression Standard , 1996 .

[13]  Alec Wolman,et al.  MCDNN: An Approximation-Based Execution Framework for Deep Stream Processing Under Resource Constraints , 2016, MobiSys.

[14]  Song Han,et al.  Deep Compression: Compressing Deep Neural Network with Pruning, Trained Quantization and Huffman Coding , 2015, ICLR.

[15]  Nicholas D. Lane,et al.  DeepEye: Resource Efficient Local Execution of Multiple Deep Vision Models using Wearable Commodity Hardware , 2017, MobiSys.

[16]  Kin K. Leung,et al.  When Edge Meets Learning: Adaptive Control for Resource-Constrained Distributed Machine Learning , 2018, IEEE INFOCOM 2018 - IEEE Conference on Computer Communications.

[17]  Song Han,et al.  Efficient Sparse-Winograd Convolutional Neural Networks , 2018, ICLR.

[18]  Tarak Gandhi,et al.  Multi-camera Based Traffic Flow Characterization & Classification , 2007, 2007 IEEE Intelligent Transportation Systems Conference.

[19]  Wei Liu,et al.  SSD: Single Shot MultiBox Detector , 2015, ECCV.

[20]  Jean-Michel Morel,et al.  ASIFT: A New Framework for Fully Affine Invariant Image Comparison , 2009, SIAM J. Imaging Sci..

[21]  Tyler Highlander,et al.  Very Efficient Training of Convolutional Neural Networks using Fast Fourier Transform and Overlap-and-Add , 2016, BMVC.

[22]  Pietro Perona,et al.  Microsoft COCO: Common Objects in Context , 2014, ECCV.

[23]  Aakanksha Chowdhery,et al.  The Design and Implementation of a Wireless Video Surveillance System , 2015, MobiCom.

[24]  Takeo Kanade,et al.  Algorithms for cooperative multisensor surveillance , 2001, Proc. IEEE.

[25]  Byung-Geun Lee,et al.  Background subtraction based on Gaussian mixture models using color and depth information , 2014, The 2014 International Conference on Control, Automation and Information Sciences (ICCAIS 2014).

[26]  Rajesh Krishna Balan,et al.  DeepMon: Mobile GPU-based Deep Learning Framework for Continuous Vision Applications , 2017, MobiSys.