Modern video analytics have developed “Search Engine” capability. Think typing some search criteria into Google and being presented with an answer. The software will sort through hours of video with ease, to quickly locate a specific person or vehicle of interest by entering physical descriptions, uploading a photo, or by finding an example within recorded video.
This technique is known as appearance search. No more will the operator be sat, trawling through hours of recorded footage. The technique provides significant improvements to incident response times and enhances forensic investigations by helping operators to compile robust video evidence, create a powerful narrative of events, and reveal a vehicle or individual’s route or last-known location.
Operators can now search for a person of interest by selecting certain from a set of variables including physical description, clothing colour, gender or age.
Where IP technology has been adopted across multiple locations, operators can even initiate a search for a person or vehicle of interest across one site, then continue the search for the same subject of interest by seamlessly transitioning from one site to the next site. Therefore, where operators are looking to monitor an individuals’ behavior across multiple locations, this is now a reality.
We’re often asked what happens if the person has changed clothing between visits. Will it still work? The answer is yes. By incorporating the unique characteristics of a person’s face enables the Appearance Search technology to understand that it is searching for the same person, even if items such as their clothing change over time.
In order for the search to be effective, of course there are some technical requirements about the environment and the way the cameras are configured. These are listed below:-
- For optimal performance, the camera needs to capture full-body and face images.
- For full-body images, position the camera to capture the following number of pixels on a person's full body - 72 pixels per meter (22 pixels per foot) based on a 2.0 MP resolution.
- To include face images to enhance a search, position the camera to collect predominantly front-facing images that use the following number of pixels on a person's full body:
- Minimum — 171 pixels per meter (51 pixels per foot) based on 2.0 MP resolution. This translates to approximately 30 pixels on the width of a face.
- Recommended — 286 pixels per meter (86 pixels per foot) based on 2.0 MP resolution. This translates to approximately 50 pixels on the width of a face.
If the camera cannot capture good face images, the search is performed on full-body images only.
Avoid having mirrors and other reflective surfaces like shiny floors and ceilings in the field of view. Reflections may cause additional false detections.
So, if your security control still operates with frustrated security officers manually reviewing hours of footage hoping to get lucky and identify a potential offender, it is time to think again.