Geovision Facial Detection

Geovision’s facial detection feature is a handy tool now but could become the best thing since sliced bread with the addition of biometric recognition.

As usual, the highest technological advancements in video analytics come down from military or ultra high-end (read ultra high cost) systems.  Geovision now supports a range of analytics as an inclusive part of it’s software.  This applies to the DVR (Digital Video Recorder) cards that sit inside a computer and take images from analogue cameras to the total NVR (Network Video Recorder) software that runs just network cameras.

Their range of analytics includes, facial detection, objects missing, crowd detection, scene stabilization, images enhancements, people counting, Pan Tilt and Zoom object tracking and a few more to boot.  Not all of these can be applied to smaller businesses but some can be invaluable.

The facial detection function is really smart and we have just activated it for 2 cameras at a busy nightclub in Copenhagen.

How it Works

Setting up the camera is all important.  In order for the system to detect faces correctly, that area should be fairly well illuminated and the faces should fill approximately 10% of the screen.  The camera we used was a Geovision day/night IP camera with a 1.3 megapixel resolution.  Here, in my view, the higher resolution really comes into its own. The camera has a vari-focal lens which means that it can be zoomed into the area you want to monitor and we chose a narrow space where everyone had to pass through on exiting.  For the shot of when the customers enter we chose an existing lower resolution analogue camera with 480 TVL (Television Lines).

The results are fantastic.  Every single face that passes by is detected and snapped into a log file.  You can see a mug shot view of all the faces that exited the club in half hour periods.  If you double click on a face it will take you directly to the video sequence (paused) when that face was detected.  the doormen have a busy job some nights with pickpockets or troublemakers.  Sometimes it can take forever to find the point where that person came in or out.  Now, although there are a lot of faces still, the task is easier and much more refined.

The Geovision day/night camera gives excellent pictures in color when there is enough light, then will change to black and white when the light level reduces. Provided there is some light, a good sharp facial picture can be extracted.  If two people are walking together it will extract both faces.

Now, I am musing over what can be done with this feature if the faces could be compared to a database.   Using my nightclub as an example, banned customers who try to come in again would be instantly flagged.  Employees entering the building can be registered as to when they start work.  There are many applications.

So it can’t be too long before those faces can be compared to a database. Come on Geovision! Pull your finger out.