Photo Courtesy Museum of Jewish Heritage
Photo Courtesy Museum of Jewish Heritage Photo Courtesy Museum of Jewish Heritage
face recognition technology
Although face recognition technology is far from perfect, it's showing definite progress. (Photos Courtesy 3VR) Although face recognition technology is far from perfect.
The ability to search by color can save significant time in video analysis.
The ability to search by color can save significant time in video analysis. (Photos Courtesy 3VR) The ability to search by color can save significant time in video analysis.
FEATURED IN INVESTIGATION
Multiple uses for intelligent video have been around for years. Red light cameras are now commonplace, and license plate recognition (LPR) is increasingly the standard for law enforcement.
Some agencies have successfully integrated different technologies that, when paired with video, complement one another.
The Los Angeles County Sheriff s Department, for example, uses ShotSpotter in tandem with a video surveillance system and LPR. In some areas, ShotSpotter identifies the location of gunshots and the camera and automatically pans to that location. This technology allows Lynnwood Substation deputies assigned to dispatch to view an incident and pass the information on to responding officers. Sometimes, street cops have a way of avoiding or ignoring new technology, but the deputies I spoke with during my visit were impressed with the technology and think it s effective.
Intelligent video is being used in New York City in a different manner. In the Wall Street area, the NYPD uses intelligent video to aid officers in conventional public safety, as well as to combat potential terrorism.
I visited the NYPD in May and toured its system. The NYPD effectively uses motion direction detection to send alerts when someone is going the wrong direction or attempting to enter sensitive locations where people should only be exiting. Intelligent video is used to identify abandoned articles in specified areas to alert law enforcement to the presence of a possible threat, such as a bomb. If an item is present longer than a specified time, it sends an alert to the command center. These tools can be effective in protecting areas and landmarks under heightened threat. This may also assist with resource allocation and allow fewer officers to cover much larger areas.
In Clovis, Calif., where I work, we re attempting to use facial recognition software to identify criminals entering our lobby. Although the jury is still out on that for Clovis, Facebook is using the same digital video recorder (DVR) vendor and has had significant success.
Where Does This Take Us?
Intelligent video has multiple uses, and, as with any technology, the advances are continuing. The use of megapixel cameras will help change the playing field for video. It s advantageous over the use of other fixed cameras and, in some cases, pan-tilt and zoom (PTZ) cameras.
We almost solely use PTZ cameras for our video policing. They re effective when an operator is at the controls and directs the camera to capture the event as it s occurring. However, on multiple occasions, we ve had crimes that have occurred where we have a camera located and the unmanned PTZ is pointed in a direction that doesn t capture the event. Although our dispatchers and officers routinely use the cameras, every camera cannot be manned at all times.
Imagine this:You ve paired a megapixel camera with the unmanned PTZ. Something happens, and you want to retrieve archived video. Depending on where it s pointed, the PTZ may or may not provide useful archived video, but the megapixel camera will most likely give you a usable image.
Imagine a recorded image that at first glance provides very little detail and is virtually valueless. Now imagine possessing the capability to zoom in on this recorded image, bringing up intricate detail with startling clarity. That s what a megapixel camera has the potential to provide. Most users I m familiar with use cameras in the 2 4 megapixel range, but cameras with significantly greater capability are readily available.
What Does This Mean?
The megapixel camera provides more data in wider camera views, which translates to more data to search for whatever it is that you re looking for. For example, take the case of a robbery suspect who you believe traveled through an area where you have a camera. The only description you have is a white male in his 30s wearing a red shirt. Using the right software, you can perform a search looking for the color red and can isolate the instances where red occurs in the search area you ve identified. Use the same scenario for a blue vehicle that you suspect was involved in a burglary. The examples go on and on.
Among the drawbacks for megapixel cameras are lighting issues and the amount of memory required to archive data. In Clovis, our biggest obstacle for facial recognition resulted from ineffective, wide dynamic range capability of our megapixel camera. This issue arises when someone is backlit so that it washes out the image you want to see. With an analog camera, the ability exists to analyze the image and darken the pixels that are too light and lighten the pixels that are too dark to create a usable image. Long ago, we experienced this in our pedestrian tunnels and overcame it by installing an analog camera with wide-dynamic-range capability.
In our facial recognition efforts, we were attempting to capture the faces of people using a megapixel camera for later comparison against our mugshot database. Our only available option was to direct the camera toward the front entrance, which introduced the backlighting problem outlined above. When we were attempting to set up this part of our system, specific technical criteria had to be met that limited the ability to deploy the technology. Because of this and recent budget limitations, Clovis facial recognition system isn t working yet.
DVR s Effectiveness
We share a common DVR vendor, 3VR (www.3vr.com), with Facebook. The Facebook IT staff made a site visit to Clovis as they were deploying their facial recognition software and encountered similar problems with backlighting. They deployed their system by using 2 megapixel cameras at chokepoints where there were no backlighting issues. Where lighting issues arose, they resorted to analog cameras even though the resolution wasn t as good.
Facebook has had significant success with its use of facial recognition software and report an 80 85% effective rate. It uses this software to track employee activity. Facebook reported having difficulties searching between DVRs, but it s working with 3VR on an enterprise solution.
Reminder:Megapixel cameras are memory hogs, and memory is expensive. Before implementing megapixel cameras you must closely examine your storage needs.
The technology has the potential to be effective, but understand it won t be 100% effective. If the intention is to create an automated system, you must realize that false alerts are going to generate problems.Example: Although Facebook s success rate is high, consider that it allows for a 15 20% failure rate. That s a lot slipping through the cracks.
Although I cite 3VR as a DVR with analytic capability, most of Clovis camera technology is Pelco (www.pelco.com), and its equipment possesses analytic means as well.
There are multiple uses for video analytics not addressed in this article, but suffice it to say that this technology can be efficient, and it will keep getting better.