When was the last time you visited a business that didn’t have CCTV cameras mounted around the facility? Retailers have them; schools and universities have them; manufacturing operations, public facilities, and even small corner stores have cameras these days. While CCTV was at first a security measure during off-hours, today those cameras are likely in operation 24/7. With image feeds from the cameras and computer vision-based analytics, CCTV systems can help operators manage and grow their businesses by providing insight from what the cameras capture. Besides providing security, here are some examples of how cameras are helping businesses:
Retail – Recognize and monitor customer movements around the store to guide store managers where to place displays and exhibits to get the most attraction; associate recognized customers and where they spend their time to notifications of future sales or new items; recognize customer reactions to products on display to inform buyers what products most excite customers.
Services – Monitor employee and customer interactions (including behavior and expressions) and employee and task execution for potential improvements in operational efficiencies.
Conferences/Educational Institutions – Monitor and manage attendees for presence and reactions to content and interaction with hosts and educators.
Public Facilities – Monitor people, recognize faces, track movements for attraction and response to public displays, etc., for fast response to safety issues and management of public information and advertising.
To be able to decode, inference, and analyze video streams is not an easy operation, especially in or near real-time. Full motion, full high definition (FHD) (or 4k ultra-high definition—UHD) CCTV cameras typically run at 25 to 30 frames per second (FPS), depending on the country. Inferencing and analyzing data streaming from hundreds of cameras like these every 1/25th or 1/30th of a second require a very efficient computer vision algorithm and a lot of compute that can also quickly decode the camera’s compression codec. Some computer vision developers have used expensive GPUs in their solutions to process all this data, and still not run in real-time.
Increasingly, companies are finding that their analytics solutions can keep up with camera feeds without costly GPUs and using CPU-optimized applications instead. One such company is AI Builder member, Accubits.
Optimizations Improve Inferencing Performance of CCTV Streams
Accubits develops solutions that add intelligence to what existing CCTV systems see. Their computer vision algorithms can detect people and their expressions, objects, and more. One of their products, called Emotyx, is an AI-powered real-time video analytics software that enables businesses to harness intelligent insights from CCTV videos. Among other use cases, Emotyx is tuned for automating attendance management at venues, such as academic institutions, conferences, and office buildings. For example, they have a 200+ camera system running at Abu Dhabi National Exhibition Center. The solution, however, is extendable and scalable to other applications.
When Accubits first developed their models with TensorFlow, Python, Flask, and other resources, they were training on Intel® Xeon® processors and inferencing with the help of NVIDIA GPUs alongside Intel® Core™ processors. Without the GPUs and using original code, their solution was very sluggish. The GPUs boosted inferencing and analytics time, but not fast enough for real-time or near real-time applications. With Accubits’ partnership with Intel, the AI Builders team was able to leverage key Intel technologies and hardware to optimize the application and increase performance significantly to deliver near real-time inference results and thus eliminate the need for GPUs.
To improve performance on CPUs alone, Intel worked with Accubits’ developers to:
- Leverage Intel Distribution of OpenVINO™ toolkit to accelerate deep learning inference for their models
- Use the Intel Optimizations for TensorFlow* and Intel Distribution for Python*, which included the Intel Math Kernel Library and other resources, for optimal performance
- Tune their codes with Intel VTune™ amplifier
CPU-based Inferencing Delivers Greater Value and Lowers Cost
What these optimizations mean for Accubits’ customers is notably greater scalability at lower deployment costs. Purpose-built hardware, such as GPUs, can often be cost-prohibitive for large deployments, which limits a customer’s ability to scale and derive maximum benefit from a solution. Additionally, performance boosts from the software optimizations that are now a part of Emotyx allow the solution to handle more cameras without having to add more compute hardware. Considering a typical Accubits installation is 100 cameras or more, customers can keep the inference and analytics hardware costs down, which is a huge plus.
Quickly and intelligently analyzing CCTV footage and capturing valuable, actionable insights from them helps companies stay ahead of the curve with both customers and competitors alike. Accubits has optimized their computer vision application for optimal performance and cost on mainstream hardware, to potentially lower CapEx and OpEx while providing critical insights that its customers need.
Individual privacy is the biggest concern for Accubits when it comes to developing computer vision-based solutions for the mass market. Their facial recognition technology employs neural encodings while saving an individual’s facial profile, which makes it harder to decipher identities even for the teams working on the solution.
Learn more about the Intel AI Builders program at https://builders.intel.com/ai.
Visit the solution catalog to learn more about Accubits’ market-ready solution for your enterprise.
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. No product or component can be absolutely secure.
For more complete information about performance and benchmark results, visit www.intel.com/benchmarks.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.