Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Collaborative intelligent cross-camera video analytics at edge: opportunities and challenges
Pasandi H., Nadeem T.  AIChallengeIoT 2019 (Proceedings of the First International Workshop on Challenges in Artificial Intelligence and Machine Learning for Internet of Things, New York, NY, Nov 10-13, 2019)15-18.2019.Type:Proceedings
Date Reviewed: Feb 19 2021

Monitoring cameras are now deployed on many city corners. When things happen, law enforcement units or other agencies can search the video feed from those cameras, either in real time or afterwards. The current approach is to send all these tasks to the cloud and send back the results afterwards. Although quite simple, this approach has several problems:

Especially for real-time video analysis, [it] often results in high bandwidth consumption, higher latency, reliability issues, and privacy concerns. Therefore, the high computation and storage requirements of [sending video to the cloud] disrupt their usefulness for local video processing applications in low-cost devices.

In this paper, the authors propose “a new paradigm for collaborative intelligent cross-camera video analytics at the edge of the network” that will “lower energy consumption, bandwidth overheads, and latency, as well as provide higher accuracy and ensure [more] privacy by leveraging knowledge sharing and spatiotemporal correlations among cameras.” Every aspect of the paradigm is covered, including privacy and adversarial attacks.

The first step is to build a smarter model. Cameras in different places have different characteristics. For example, human behavior captured by a camera facing a parking lot would be very different than behavior captured by a nightclub-facing camera. Thus, “it would be more accurate to build a customized model for each camera based on its local data,” rather than using the same model in the cloud.

Next, when two adjacent cameras have an overlapping covered area, the same area should be analyzed only once. It also provides load balancing opportunities between adjacent cameras.

The third approach is sharing high-quality captures with those of lower quality. It enhances the correctness of the tasks. Also, after an object is identified in one camera, when the same object moves to the next monitoring camera it should not go through the identification process again.

Although the ideas presented are good, implementation is not easy considering the various factors facing street cameras. Things will be even more complicated if malicious and broken cameras are present.

Reviewer:  R. S. Chang Review #: CR147194 (2107-0188)
Bookmark and Share
  Featured Reviewer  
 
Computer Vision (I.5.4 ... )
 
 
Architectures (H.5.4 ... )
 
 
Distributed Architectures (C.1.4 ... )
 
 
Network Architecture And Design (C.2.1 )
 
Would you recommend this review?
yes
no
Other reviews under "Computer Vision": Date
Machine vision
Vernon D., Prentice-Hall, Inc., Upper Saddle River, NJ, 1991. Type: Book (9780135433980)
Oct 1 1992
The perception of multiple objects
Mozer M., MIT Press, Cambridge, MA, 1991. Type: Book (9780262132701)
Mar 1 1993
Computer vision, models and inspection
Marshall A., Martin R., World Scientific Publishing Co., Inc., River Edge, NJ, 1992. Type: Book (9789810207724)
Jun 1 1993
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy