Expanding AI-Powered Search with Reverse Image Analysis

When it comes to solving security incidents, organizations often contend with incomplete or missing information. Perhaps an eyewitness only identified a suspect’s shirt color or a cell phone camera only captured a portion of a suspicious vehicle. Verkada’s search capabilities use the latest in AI and computer vision to deliver results in situations where descriptive information is otherwise limited. Take, for example, the first iteration of our AI-powered search: if a suspect is spotted in a denim jacket without other identifying attributes, organizations can search with everyday language (“person wearing denim jacket”) to return footage of all people wearing denim jackets in a matter of seconds. There are, however, countless other details of an individual, vehicle, and overall scene – beyond what we can describe in words – that can contribute to an even more precise and complete investigation. Our latest improvement to AI-powered search – incorporating reverse image search – analyzes photos relevant to investigations to help overcome these information constraints. 

Now, in addition to using freeform text with AI-powered search, organizations can upload a photo of a person, vehicle, or object to Command to obtain relevant search results. Akin to Google’s Reverse Image Search, this new functionality pulls multiple aspects of an image, like shapes, pixels, patterns, and other granular details, then delivers the likeliest set of matches from one’s camera fleet. While the results of this search will display thumbnails of people and vehicles, [1] organizations can use it to see if non-person and non-vehicle objects were carried, held, worn, or attached to a person or vehicle. As illustrated in the example below, organizations can upload a photo of a standalone bag to retrieve results of any individual(s) who may have held (and left behind) a specific bag – a use case with clear security benefits:

This new search capability also proves useful when faces are not visible. Because it does not rely on facial recognition to return results, reverse image search can be used to help identify suspects of interest even when their faces are partially visible or not visible at all – offering another complementary tool in Verkada’s search arsenal to identify suspects in seconds. Figure 2 below illustrates this capability:

With reverse image search, we’re continuing to lead the video security industry with differentiating technology that help organizations conduct efficient, comprehensive investigations. This latest AI-powered search enhancement will be available beginning on November 14th, 2024.

[1] Consistent with AI-powered search, results leverage “hyperzooms,” which are highly-detailed segments of a video frame that contain images of a person or a vehicle only, typically the most important parts of an investigation. If an uploaded image does not include a person or vehicle (i.e., a standalone object), results will be directed to “hyperzooms” of people and vehicles that include the object (e.g., if uploading a standalone suitcase, the search will return results of a person holding a suitcase or a suitcase attached to a vehicle).