The cloud and deep learning are two technologies on the verge of transforming the video surveillance market. You often hear the two technologies mentioned as separate trends, but lately, I have been hearing about how interdependent they are.
Economies of scale provided by the Cloud (Software as a Service, or SaaS) are making the sophisticated capabilities of deep learning affordable to a wider audience. Meanwhile, deep learning augments cloud systems with capabilities that may not be available, or affordable, in on-premise systems, thus accelerating the broader move to cloud systems.
A new era of autonomous video
One company that is combining the cloud and deep learning in an end-to-end system is Umbo Computer Vision, an “autonomous video security technology” company built on its founders’ expertise in deep learning and machine vision. The company’s cameras are equipped with basic video analytics, and they are tied into a cloud infrastructure that provides additional deep learning algorithms. The cameras provide computer vision pre-processing, with the bulk of detailed analysis happening in a neural network in the cloud.One customer benefits from another customer and all the knowledge is aggregated together"
“Having a SaaS model gives companies like ours the resources to improve the deep learning model,” says Shawn Guan, CEO and co-founder of Umbo Computer Vision. “We can make more accurate systems that scale better and faster. SaaS enables vendors to do something great with deep learning. You don’t have to redo it for everybody. One customer benefits from another customer and all the knowledge is aggregated together.”
Adapting to a changing landscape
Umbo’s deep learning analytics provide higher accuracy, says Guan. For example, the system can provide accurate alerts on people climbing a wall, and it can accommodate for elements such as shadows and adverse weather conditions. The cameras can also provide occupancy information from a video view of a room; unlike people-counting applications, the camera does not need to be mounted above the door.
The algorithms can detect loitering and can track a subject even if he or she disappears briefly from view (such as walking behind a building). Umbo developed the algorithms from scratch using the company’s deep roots in technology. Guan says the company was “built on data and using data.”
 |
| The company’s cameras are equipped with basic video analytics, and they are tied into a cloud infrastructure that provides additional deep learning algorithms |
The cloud serves as a media gateway
“Cybersecurity is ensured,” says Guan. The cloud provides built-in authentication with the camera; user names and passwords are handled in the cloud, so there is no manual reset of the password at the device level. The cloud serves as a media gateway, and two-way communication between the camera and the cloud addresses how much data is streamed.
Data in the camera has 256-bit encryption, and there is a secure “tunnel” through which data moves in the Internet. In the cloud, the host provider, such as Google, Microsoft or Amazon, protects the data to their own high standards. Each camera has industrial grade SD providing 30 days’ storage of full frame rate, 30fps video at 1080p, with backup in the cloud. In effect, the camera operates in-lieu of a network video recorder.The end-to-end system is designed to handle an essentially-unlimited number of cameras and can also be integrated with other products
Handling unlimited cameras
Customers interface with the system through the cloud’s software user experience. The end-to-end system is designed to handle an essentially-unlimited number of cameras and can also be integrated with other products. Umbo targets the enterprise market and has around two-hundred enterprise customers in the United States - their biggest market - as well as offices in London and Taipei, and customers around the world. The cost structure involves an annual payment per camera, between $280 and $600 depending on features/capabilities.
Umbo is introducing a new bullet camera with a zoom lens for exterior applications, incorporating a higher-end Sony sensor for good low-light performance and wide dynamic range (WDR). The camera’s modular design enables the addition of features such as longer-range infrared – increased from 100 to 160 feet – white-light LED to provide a colour image, and greater deterrence, at night, and a long-range microphone for audio. The company also offers a fixed-lens dome camera for indoor applications.
