There are a number of different people counting sensors available on the market. Single lens (Mono lens / 2D) sensors use a technique called ‘background subtraction’ to detect movement when groups of pixels change colour. In scenarios where there is direct sunlight or shadows caused by lighting, the background subtraction technique struggles to maintain accuracy. Some single lens cameras try to use AI to improve counts by identifying patterns of people from the background, however, the main problem is, you’re still dealing with significant exposure issues and reliant on a training set that cannot cover every eventuality.

As an example, the images alongside show the difference in lighting from morning to afternoon. Shadows and overexposed areas in the afternoon cause a particularly challenging environment for a single lens camera. While most sensors will struggle to determine the number of people accurately in these scenario’s, this is not the case with Accu-Count.

Rather than using pattern recognition, Accu-Count sensors use Dual-lens stereoscopic (left and right) depth perception technology and a non-visible infra-red pattern projector to build a 3D model of the entrance and determine and track the peak of each target as follows:

• The left and right cameras capture the scene. The image alongside shows 6 people walking through the entrance.
• Depth values are calculated for each pixel by correlating points from the left image to the right image.
• Depth pixel values are processed to generate a depth frame. This is enhanced by the infra-red pattern in images with low texture and light. The depth frame alongside shows the peaks of all the objects moving in its field of view.
• Subsequent depth frames create a depth video stream which is used to track and count the peak of objects (a person’s head) and not the whole object.