Why count ties on your exhibition stand?

GPS, translators, search engines… Today, we can easily imagine intelligent applications in our daily lives and yet artificial intelligence algorithms have many other fields of application, for example counting the number of ties passed on your exhibition stand.

Artificial intelligence is now part of our daily lives. Most of us are conscious users, informed or not about intelligent solutions. To highlight the possibilities offered by AI techniques associated with vision, we have developed an artificial vision demonstrator which makes it possible to quantify the number of visitors to a stand during a trade show and to detect the presence of specific objects.

Industrial vision on your stand

The demonstrator developed is made up of two parts:

  • the PC which embeds the calculation code and analyses the images from the webcam in real time,
  • the graphics interface which reports on the captured image and the associated statistics.

Who really visited your stand?

To assess stand footfall, we need to count the number of people who come. To do this, this demonstrator uses the latest powerful computer vision algorithms based on “Deep Learning” to perform object segmentation.

The principle is quite simple:

  • within a video stream, we freeze the scene every 2 s to extract an image,
  • on this image, a segmentation is carried out for a set of objects which are then counted.

Démonstrateur Vision

Ties, Smartphones, bags, etc.…

For this demonstrator, the objects we decided to identify are ties, Smartphones, bags (handbags, backpacks, shopping bags), and people. To do this, we chose the “Faster R-CNN” algorithm which makes it possible to segment the objects (isolate regions in the image containing the different objects that we are looking for). For optimal operation, this algorithm needs high-performance hardware resources, in order to obtain results within a reasonable time.

Detection every 2 seconds

To measure the attendance at a trade show effectively, we found it sufficient to have object detection updates every 2 s by exploiting the performance of the graphics card. Indeed, for most “Deep Learning” algorithms, accuracy comes from learning (training) on ​​a large number of references requiring significant hardware resources to progress quickly. Moreover, the update frequency of the segmentation, within the video stream, requires resources to have a respectable refresh rate in the graphics interface.

Deep learning and data

In our case, the objects we chose to recognise have been learnt out of millions of items in order to be able to consider:

  • the context: for example, a tie is better recognised if it is tied or a person can be detected from his hand,
  • orientation changes,
  • lighting changes.

Statistical analyses

The graphics interface, developed in Python, presents attendance statistics (number of people) over the last 15 minutes to provide “live” information and a curve over the day which is compared to the day before.

To obtain realistic results, with each new segmentation performed by the algorithm, only the new people in the image are counted (Demonstration video)

Algorithm and count

For example, if between two images the number of people changes from 2 to 4, the count will be incremented by 2. However, if the number of people changes from 4 to 2, no increment is made. It is assumed that 2 people are no longer on the stand since they have moved away from the camera’s field. Moreover, we integrated detection filtering into our algorithm in order to count only people in a close visual field. If people are detected far from the stand, they are not counted.

Algorithm and ethical choice

For our demonstrator, we would like to emphasise that no tracking of people is performed. This could be performed in applications requiring more precision in order to avoid counting the same person several times when they return to the stand. The choices we made seem sufficient in our application since the goal is simply to present an application scenario, by displaying a monitoring of attendance every hour.

Defective product, video surveillance, counting

As part of our application, we used a graphics card to speed up the image processing rate. This is not required if the application does not show rapid changes: slow movement, infrequent new detection, etc.

The principle highlighted by this demonstrator, artificial intelligence associated with computer vision, can be used to count any “object”. This approach can easily be exported to production lines in order to estimate the production rate, detect defective products or even carry out video surveillance.

From ties to cows

We could also count the number of cows easily or manage the flow in the milking parlour. The only constraint, in exploiting the same principles as those of the demonstrator, is the installation of a camera and diversified, realistic annotation in order to make appropriate detections.

Syntyche Gbéhounou, April 2021.

Share this post: