printer friendly version

Optimising deep learning for visual surveillance

Advances in vision technology, such as enhanced detection accuracy and deep learning flexibility are transforming the surveillance market. Across industries, businesses and organisations are looking to harness these advances to help improve their security, safety, and operational effectiveness.

However, there are key performance requirements that must be met before smart vision solutions can drive effective and accurate results for customers. To truly fulfil the promise of smart vision, solutions must:

• Integrate with current and legacy systems to serve as the foundation for business processes.

• Maintain high levels of accuracy and reliability to unlock actionable data, meet standards of evidence, and comply with privacy requirements.

• Scale to capture and analyse vast amounts of visual data across distributed environments such as campuses, roadways, and cities.

The Intel Vision Products portfolio

The Intel Vision Products portfolio is comprised of silicon, software tools, deep learning frameworks, and libraries that are positioned for the next generation of AI. Intel Vision Products (IVP) are helping put your data to work from the edge to the cloud, so you can act in real time, make decisions faster, and implement new operational strategies to drive immediate results.

At the hardware level, Intel boasts a comprehensive selection of acceleration silicon. Intel CPUs, CPUs with integrated graphics, and Intel Vision Accelerator Design Products based on Intel Movidius VPUs and Intel FPGAs (field programmable gate arrays) help deliver highly accurate vision analytics performance and compute efficiency. (An FPGA is an integrated circuit designed to be configured by a customer or a designer after manufacturing – hence the term ‘field-programmable’ – Wikipedia.)

Intel also offers an array of software tools, including the Intel Distribution of Open Visual Inference and Neural Network Optimisation (OpenVINO) toolkit, for accelerating the development and integration of intelligent vision solutions and capabilities at scale. This end-to-end suite helps scale and integrate vision capabilities across your entire end-to-end infrastructure – whether for premises, campuses, or city-wide applications.

Actionable intelligence

ISS (Intelligent Security Systems) takes a smart and innovative approach to applying deep learning technologies, embedding purpose-built, task-specific, and edge-driven neural networks into its complete hardware-to-software solutions. Neural network-based algorithms allow ISS to bring current solutions to a new quality level and build new solutions with capabilities not reachable before – including key smart cities use cases such as licence plate, cargo, and facial recognition, and more.

SecurOS is ISS’s suite of VMS offerings for on-premise and cloud-based video analytics. SecurOS is designed to scale with customers as their deployments and surveillance needs grow, comprising a suite of offerings for multiple intelligent video applications. SecurOS-based systems easily integrate with most legacy digital security and surveillance systems, and ISS can provide local system integration and support to streamline the deployment process.

The Intel Distribution of OpenVINO toolkit is helping ISS deliver neural network-enabled surveillance products to support wide-ranging applications. The OpenVINO toolkit optimised ISS’s deep learning neural networks for Intel’s suite of compact, low power, high performance CPUs and CPUs with integrated graphics. For high powered, dedicated edge workloads, the toolkit will bring ISS’s neural networks to its latest Intel Vision Accelerator Design Products based on the Intel FPGAs and Intel Movidius VPUs.

By leveraging the toolkit, users can accelerate computer vision performance, shorten vision solution development, and streamline deep learning inference and deployment.

ISS Africa, headed up by Eugene Kayat, will be focusing on Africa-wide expansion in the coming year.

Share this article:

Further reading: