Edge based visual intelligence

Until recently the design of high quality security cameras have always been at the forefront of the electronics industry. Digital pixel technologies led to unprecedented gains in image quality and framerates, boosting video performance. Surveillance cameras became better observers, but still functioned like portholes; useful only when someone looks through them. Today’s innovations concentrate on device intelligence by giving a camera onboard intelligence, empowering edge computing.

Cameras with visual intelligence

In order to be useful, data needs interpretation and images require analysis to become meaningful with its content. A camera with onboard intelligence interprets image data from the sensor and gets an understanding of the world in its view. It can decide to output a message to a connected system or device to trigger alarms or automate processes. It can be part of an integrated system solution or operate standalone in a local network. At the heart of this development is artificial intelligence.

Mimicking the human brain

Artificial intelligence is a computing system that mimics natural intelligence. A digital mind can be trained to perform pattern recognition and image analysis. These networks, tailored to perform specific visual recognition tasks, are implemented and executed on the camera’s hardware.

To grasp the idea of a neural network, imagine what happens in your brain, when a car is driving towards you. Visual data coming from your eyes flows through your brain. Specialized layers of neurons pick out the cars visual features and reassemble the final image. The same way a deep neural network breaks down camera images. Raw data is fed through several layers of trained neurons each extracting image features. The first layers finds rough edges, the next layers add up to refinement and complexity. A final layer combines all details and comes to the final conclusion; It is a car and it drives towards me!

Edge based computing

Edge computing occurs directly on the device that contains the visual sensors, without shifting the compressed data anywhere else. Only crucial information – meta data – is transmitted, reducing bandwidth usage and costs by a significant degree. Security is improved as usage of the public Internet is trivial. By bringing image processing closer to the source, the generation of data-driven decisions is more accurate, faster and improves system response time. Where cloud computing needs 24/7 Internet access, an IoT sensor works even without the Internet.

Making alerts and alarms meaningful

Most cameras can detect motion in its scene, but is this want you need? If a security operating center receives alarms when a cat moves over the premises, centralists will become lesser focused. What if an intelligent camera can detect objects like persons or vehicles? Then the centralist receives a true alarm to follow up, making security services far more efficient than it ever was.

With computational power growing and AI algorithms evolving, device intelligence develops to a new level. A visual IoT sensor now has a local brain, processing images and being capable of making standalone decisions. For us it is a challenge and a privilege to create intelligent products, to see them at work in the field and to satisfy or surpass the user’s needs.