NEC takes real-time video analytics to the edge

Japan's NEC said it's on the cusp of launching 'gradual deep learning-based object detection technology' to aid real-time video analytics.

Ken Wieland, contributing editor

December 1, 2021

3 Min Read
NEC takes real-time video analytics to the edge

Japan's NEC said it's on the cusp of launching "gradual deep learning-based object detection technology" to aid real-time video analytics.

The new-fangled software, claimed NEC, enables up to eight times the processing speed of object detection for large volumes of images, "even on edge devices with limited processing capacity." NEC seemed pretty excited about the whole thing. The technology, admitted the firm, still needs some R&D attention but that didn't stop it from trailing a commercial launch, expected in NEC's fiscal 2022 (which ends March 31).

Figure 1: (Source: NEC) (Source: NEC)

NEC highlighted a wide range of applications for "real-time" video analytics, including analysis of camera images of vehicles at intersections, optimizing traffic control, and license plate detection. Analyzing camera images of stores and warehouses, either to detect intrusion or optimize facility management, were flagged as other useful apps.

Yet NEC further pointed out – to set the stage for its "high-precision object detection acceleration technology for edge equipment" – the difficulties of doing real-time video analytics. In an ideal world, said NEC, processing of images should be done near a camera but that's apparently not possible using conventional methods.

Figure 2: (Source: NEC) (Source: NEC)

For one thing GPU processors used in high-performance servers are not available. Moreover, added NEC, cooling is difficult at the edge. Electricity consumption is also restricted. The upshot is that edge processing capacity is necessarily constrained.

Enter deep-learning object detection

NEC makes the case that highly accurate object detection AI models have large amounts of operations, which make it difficult for edge devices to process lots of images. To make matters worse, if the number of operations for a high-speed object detection AI model is reduced, the accuracy declines and – according to NEC -- recognition accuracy requirements for image analysis can't be met.

Want to know more about the Edge? Check out our dedicated edge content channel here on
Light Reading.

Cue object detection software, said NEC, which utilizes deep learning to find the object required for analysis from camera images. According to NEC, the newly developed approach enables efficient, high-speed, and high-precision detection of subjects from large amounts of images, even in an edge device with limited processing capacity.

It also purportedly enables simultaneous processing of images from multiple cameras in real time.

Related posts:

— Ken Wieland, contributing editor, special to Light Reading

Read more about:


About the Author(s)

Ken Wieland

contributing editor

Ken Wieland has been a telecoms journalist and editor for more than 15 years. That includes an eight-year stint as editor of Telecommunications magazine (international edition), three years as editor of Asian Communications, and nearly two years at Informa Telecoms & Media, specialising in mobile broadband. As a freelance telecoms writer Ken has written various industry reports for The Economist Group.

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like