Dive Brief:

• Verkada has used NVIDIA tools to add what it calls a more robust image search capability to its AI-assisted security camera systems, it says.
• The company used model simulations from NVIDIA’s Cosmos program and computing architecture from NVIDIA’s Physical AI Data Factory to improve how its camera AI systems train and make inferences, it reported last week.
• “Working with NVIDIA supercharges what we've spent nearly a decade building,” Verkada CEO Filip Kaliszan said in a statement: “AI that keeps students safe in schools, protects workers on factory floors, helps retailers prevent theft, and enables organizations to operate more efficiently."

Dive Insight:

In an AI-based camera system, image search matches what users are seeing with what cameras have picked up in the past. If a camera picks up a person who’s behaving suspiciously, for example, it can match that person with similar images from the past.  

“You can go back [to the video feed] and say, ‘Give me a person with a red hat that was here within the last six months,’” Michael Evanoff, Verkada’s global chief security officer and strategic advisor, told Facilities Dive last year. “Within seconds, it will pop up [other images of] a person with a red hat.” 

At an arena or an office, for example, people who have a history of disruptive behavior can be entered into the system and flagged if the camera sees them again. “The system alerts you on your phone,” Evanoff said. 

The AI training and inferencing from the NVIDIA collaboration is intended to improve the camera system’s ability to make a match, among other things, the company says.   

Using Cosmos, NVIDIA says in a technical paper, “Physical AI agents acquire two fundamentally coupled capabilities: understanding and generation.”

NVIDIA says its Physical AI Data Factory works with Cosmos by extrapolating from limited data sets to create large-scale data sets that Cosmos can train on, including “rare edge cases and long-tail scenarios that are expensive, time-consuming and often impractical to capture in the real world.”  

Verkada says it’s seen an improvement in its image search since it started using the NVIDIA tools. “The mean average precision (mAP) of … AI-powered search [has improved] by 68% for spatial-temporal understanding, delivering faster, more accurate, and more robust search capabilities,” it said in the press release.

The company also is using the tools to make improvements to its search capabilities to better identify what it calls complex, unstructured real-world scenarios, “from identifying health and safety incidents on a manufacturing floor to detecting shrinkage in retail environments,” it stated.  

NVIDIA has made an equity investment in Verkada as well, the company said, “following a strategic investment from Alphabet's CapitalG at the end of last year.”