Face & Person Detection
Facial recognition is hot.
Most security cameras today have enough processing power to analyze faces and compare them to faces in a database.
Sometimes, however, it is not necessary or (for privacy reasons) desirable to recognize someone with your security cameras, but you just want to know that a person is in the picture and generate an event or alarm based on this observation.
This is referred to as face or person detection. Facial recognition is useful if you want to track specific people or as part of your access control system.
Facial recognition starts by detecting a face: an oval shape containing two eyes, a nose and a mouth.
After that, the algorithm will compare thousands of measurement points in the face with the same measurement points of the faces in the database.
This step takes a lot of processor capacity.
If it is not necessary to recognize people or if you do not want to do so for privacy reasons, you can also use the software in the security camera to detect faces and/or people.
In this way, you can, for example, quickly filter video recordings, generate an event or alarm, or limit the storage of images by only recording when a face or person is in the frame.
What are the limitations?
Depending on why you want to use face detection, this technology also has limitations.
This is because a security camera generates two-dimensional images.
This means that the camera can also be fooled with photos and drawings of a face.
If face detection is only used to start a recording or to quickly retrieve images, this doesn’t matter.
At most, you will get a few more images than intended.
Another limitation is that covered faces are not recognized.
Even headgear and scarves can cover the face in such a way that the detection rate is too low to generate an event or alarm.
Face detection alone is therefore not suitable for alarms.
There are simply too many false positives and false negatives.
Ideally, you only want to have ‘true positives’.
Combined Detection Increases Reliability
By combining face detection with person detection, reliability increases considerably.
Person detection is an algorithm that looks at whether the moving object is a human.
It’s a more complex algorithm than the one for face detection, but it uses less processor power than facial recognition.
Moreover, privacy is not at stake.
The camera software can be adjusted in such a way that when both a person and a face are detected, there is an event.
Detection of a person, without face detection, is good for a higher class event.
This combination indicates that someone is shielding or covering their face, which can indicate a suspicious situation.
Face detection without person detection sounds a bit spooky, but usually indicates a face used on a billboard or car with advertising.
There is no need to give an event/alarm for this.
Conclusion
This combination of face and person detection can be a good addition to the existing surveillance system.
In many new cameras, these algorithms are already standard and it is a simple upgrade to next-level motion detection.
Moreover, thanks to developments in computing capacity and deep learning, these technologies are becoming increasingly reliable in the short term.