At ACOREL we have been working on automatic people detection and counting for more than 3 decades. Along the years this has involved using different type of technologies and sensors ranging from passive and active infrared sensors, 2D cameras with image processing technics, up to Lidar sensors and 3D stereoscopic cameras.

The detection efficiency and resulting counting accuracy has increased over the years from around 90% with passive infrared sensors up to 99% with 3D stereoscopic cameras. This massive improvement has been possible thanks to the technology of the sensors themselves, but also because of the increasing in processing power of the CPUs and GPUs

How to optimise counting accuracy ?

In the past few years, a new type of technology has made a breakthrough. This is Artificial Intelligence and Machine Learning. It is far from being a new concept in computer science. Indeed, Artificial Intelligence and Machine Learning have been existing from many years. But the increasing processing power of modern computers and new applications (like self-driving cars, face recognition, etc…) have push these technologies to the foreground.

This is especially true for Computer Vision, and this might as well be a game changer for automatic people counting and occupancy applications.

People counting with computer vision uses deep learning algorithms to detect and track individual people in the real-time video of common, inexpensive surveillance cameras. New deep learning algorithms provide high accuracy in both indoor and outdoor scenes

Data detection

Detecting and counting people using 2D cameras and image processing technics is not new, and at ACOREL we’ve been doing it with more 95% of accuracy from 20 years. It was based on complex image processing, requiered quite high processing capacities and was very sensitive to the environnement, especially oudoor installations in direct sunlight. In addition, it was quickly superseded by technologies like LIDAR and 3D cameras which provides higher accuracy at lower costs

Artificial Intelligence and Machine Learning applied to Computer Vision brings a complete different approach to people detection and counting : it is now possible to train different sort of neural networks to automatically detect any kind of object on an image or a video stream, including people, in any environment with a very good accuracy. This opens the door to realtime occupancy calculation and people counting using cost effective 2D cameras and processing units.

This technologie can be used everywhere (not only on airport), which opens up a wide range of possibilities.

If you want to find more information, click here