Science & Technology

Deep North, which uses AI to track people from camera footage, raises $ 16.7 million

Join the online game readers at GamesBeat Summit Next. It will be held from November 9th to 10th. Learn more about what’s coming next.

Deep NorthA Foster City, Calif.-based startup applying computer vision to security camera footage today announced that it has raised $ 16.7 million in a Series A-1 round. According to CEO Rohan Sanil, with the participation of Conviction Investment Partners, led by Celesta Capital and Yobi Partners, Deep North will use this money to hire and expand its services “on a large scale.”

Formerly known as Vmaxx, DeepNorth “accepts digital” for retailers in stores by improving its security system to track purchases and ensure compliance with masking rules. Claims to help protect against COVID-19. However, the company’s system, which relies on potentially flawed algorithms, raises concerns about both privacy and bias.

“Even before the pandemic caused retailers to close their doors … companies were struggling to compete with the fast-growing online consumer base,” Sanil said in a statement. “As stores reopen, retailers are adopting creative digital solutions with data-driven results-based computer vision and AI solutions to become more competitive with online retailers while at the same time COVID safe practices. Must be addressed. “

Monitoring using AI

Deep North was founded in 2016 by Sanil and Jinjun Wang, experts in multimedia signal processing, pattern recognition, computer vision and analysis. Wang, now a professor at Xi’an Jiao Tong University in Xi’an, China, was formerly a research scientist at NEC before joining Epson’s R & D department as a member of senior technical staff. Sanil has founded many companies before Deep North, including Akirra Media Systems, where Wang was once employed as a research scientist.

“In 2016, I pioneered object detection technology and promoted targeted advertising from online videos. When a major brand saw this, it identified and analyzed objects captured by theme park security camcorders. , And challenged me to create a way to sort, “Sanil told VentureBeat in an email. “My quest has unleashed the potential of CCTVs and security camcorders installed within the customer’s physical environment and has influenced the development of applying object detection and analysis in all formats of video.”

After opening offices in China and Sweden and rebranding in 2018, Deep North has expanded the availability of computer vision and video analytics products that provide object and person detection capabilities. According to the company, AI-powered, hardware-independent, real-time software is available for retailers, grocery stores, airports, drive-throughs, shopping malls, restaurants, and events.

According to Deep North, retailers, malls and restaurants in particular use the solution to analyze customer “hotspots”, seats, occupancy, dwell times, gaze directions and wait times and leverage these insights. You can then decide where to assign the clerk or kitchen. staff. Stores can predict conversions by correlating tracking data with time of day, location, marketing events, weather, and more. Shopping centers can use tenant statistics to understand trends, identify “synergies” between tenants, and optimize store placement and crossing. Tenant promotion.

“Our algorithms are trained to detect moving objects and generate rich metadata about the physical environment such as engagement, paths, and dwellings. Our inference pipeline is real-time processing. We’ve put together a camera feed and algorithms for you, “Deep North explains on its website. “”[We] It can be deployed both in the cloud and on-premises and up and running within hours. Our scalable GPU edge appliance allows businesses to bring data processing directly to the environment and transform properties into digital AI properties. Video assets never leave the premises, ensuring the highest level of security and privacy. “

In addition to these solutions, Deep North has developed products for specific use cases such as social distance and hygiene. The company offers products that monitor hand washing, estimate waiting times at airport check-in counters, detect the presence of masks, and track the status of tarmac maintenance workers.

“DeepNorth’s mask detection feature makes it easy for retailers to monitor large numbers of people and receive real-time alerts,” explains DeepNorth’s social distance product. “In addition, Deep North … monitors the schedule and scope of hygiene measures, and the total time taken for each cleanup activity … using Extensive Data from Deep North, [malls can] Create a tenant compliance scorecard to benchmark your efforts, track overall progress, and modify courses as needed. [They] You can also monitor real-time occupancy on dashboards and mobile apps to ensure that occupancy limits are being adhered to across multiple properties, both locally and across regions. “

Concerns about bias

Like most computer vision systems, Deep North has been trained with image and video datasets that show examples of people, places, and objects.Inadequate representation in these datasets can be harmful — especially considering the AI ​​field. Generally lacks a clear explanation of bias..

In a previous survey ImageNet Open Images — two large public image datasets — are centered around the United States and the euro, encoding human-like prejudices about race, ethnicity, gender, weight, and more. Models trained on these datasets Global North and Global South.. For example, the image of the groom is less accurate than the image of the groom in the United States when it comes from Ethiopia and Pakistan. Also, depending on how images of words such as “wedding” and “spice” appear in distinctly different cultures, object recognition systems can be used if many of these objects are from the Global North and Global South. You may not be able to classify them.

Bias can come from other causes, such as differences in the sun’s path between the northern and southern hemispheres and changes in the background landscape. Studies show Camera model — For example, resolution and aspect ratio — can reduce the effectiveness of classifying objects that have been trained to be detected by the algorithm.

Technology companies have traditionally introduced defective models into production. ST Technologies’ face recognition and weapon detection platform misidentifies black children at a higher rate, Frequently mistaken the handle of a broom for a gun.. Meanwhile, Wal-Mart’s AI-based and camera-based anti-shoplifting technology Have seen so farWas scrutinized on it last May Reportedly The detection rate is low.

Deep North does not disclose on its website how it trained computer vision algorithms, including whether they were used. Synthetic data (It has its own flaws) Supplements the actual dataset. The company also doesn’t say whether it takes into account users with accessibility and key mobility issues. People with disabilities, for example, may have patterns of gait or limb movements that appear to differ from the algorithms trained in healthy person footage.

In an email, Sanil claimed that Deep North “has one of the largest training datasets in the world,” derived from actual deployments and scenarios. “Our human detection and analysis algorithms are trained in over 130 million detections, thousands of camera feeds, and a variety of environmental conditions to provide our customers with accurate insights,” he said. Told. “Our automated semi-supervised training methodology helps us quickly build new machine learning models with minimal training data and human intervention.”

Privacy and controversy

The goals of products like Deep North are health, safety and analytics, but this technology could be adopted for other inhumane purposes.Many privacy experts are worried about what they do Normalize Higher levels of monitoring, collecting data on worker movements, and allowing managers to blame employees in the name of productivity.

We have also collaborated with school districts and universities in Texas, Florida, Massachusetts, and California to pilot a security system that uses AI and cameras to detect threats, which has been controversial. Deep North claims that the system, which was subsequently deprecated, works with 320p low-resolution cameras and can interpret people’s behavior while identifying objects such as unmanned bags and potential weapons.

Deep North is also working with the Transportation Security Administration, which provided a grant last March, to test biometrics, including a self-screening portal, at Detroit Metropolitan Wayne County Airport. The company provides indicators such as passenger throughput, social distance compliance, agent interaction, bottleneck zones, as well as unmanned baggage, misdirection, or restricted area occupancy. Received nearly $ 200,000 in funding for this.

“We are humble and excited that TSA can apply innovations that will help us realize our vision of improving the passenger experience and safety of the entire airport,” Sanil said in a statement. “We are committed to providing the US Department of Homeland Security and other government agencies with the best AI technology to build a safer and better homeland through continuous investment and innovation.”

In an interview with the Swedish publication Breakit, Deep North confirmed that it offers facial recognition services to some customers. And on that website, startups are touting their technology’s ability to personalize marketing material according to people’s demographics, such as gender. However, Deep North is sticking to its internal protection so that the identity of the person captured through the camera footage cannot be identified.

“There is no ability to link metadata to a single individual. In addition, Deep North does not obtain personally identifiable information (PII) and manages the integrity of all individuals with the highest possible anonymization standards. And was developed to maintain, “Sanil told TechCrunch in March 2020. All PIIs only store derived metadata that produces metrics such as the number of entries and the number of exits. DeepNorth strives to maintain compliance with all existing privacy policies, including the GDPR and the California Consumer Privacy Act. “

To date, 47 employees, Deep North, have raised $ 42.3 million in venture capital.

Venture Beat

VentureBeat’s mission is to become a Digital Town Square for technical decision makers to acquire knowledge about innovative technology and trading. Our site provides important information about data technologies and strategies to guide you when you lead your organization. We encourage you to become a member of the community and access:

  • The latest information on the subject you are interested in
  • Newsletter
  • Gated sort reader content and discounted access to valuable events such as: 2021 transformation: learn more
  • Network function etc.

Become a member

Deep North, which uses AI to track people from camera footage, raises $ 16.7 million Deep North, which uses AI to track people from camera footage, raises $ 16.7 million

Back to top button