Hello,
does anyone have experience counting toy cars with an S2 sensor? For events in Berlin, we want to use an S2 counter to count different types of toy cars in order to get children and young people excited about the topic of collecting mobility data using inexpensive counting devices. Telraam has already implemented something like this for its own trade fair stand demo.
What proportions are best suited for this? We are currently trying it out with a Lego Mindstorm model with an attached silhouette at a scale of 1:20; other scales of 1:18, 1:24, 1:43, 1:12, 1:32 and 1:64 might also be useful. We want to place the S2 counter in a house window of a Lego city (street). Counting Carrera cars would also be great.
Sorry for not getting back to you on this. How has the project gone?
Indeed, we have developed what we call the ‘diorama’ that we use on our stands when presenting how the device works. We use toy cars, bikes and other vehicles to show how the device counts them.
It should work, in principle, with most cars that are based on actual models and therefore have the right proportions to be identified as a ‘car’ (or other vehicle). It can be sensitive to having many different scales (e.g. toy buses that are the same length as toy cars and therefore scaled differently) so you may need to test it.
Once in place, you may need to reset the ROI to an appropriate (wide) view to make sure it captures the full object, depending on how close it is.
Have you had a chance to test it?
Hi Rob,
I made a first test with a toy bus (scale 1:48). Out of 10 attempts, 8 to 9 were usually counted successfully, but not as a bus but as a car. You can watch the test here. I set the ROI to 11.
Can you tell me, which toy models you have used?
I’m going on a long bike tour starting tomorrow and therefore won’t be able to continue the tests myself until May.
On July 17, the ADFC Treptow-Köpenick team held the first toy car counting event with seventh-grade children at the DLR_School_Lab in Berlin Adlershof. The research task was to find out how to conduct meaningful traffic counts and, of course, why this data is needed in the first place. The students had great fun and enthusiasm building a small LEGO city. There, they were tasked with incorporating three Telraam traffic counters, which the ADFC also uses in real-life measurements. Their imaginations knew no bounds, and some truly interesting structures emerged! This event was made possible by the generous donation of LEGO and DUPLO building materials from the LEGOLAND Discovery Centre Berlin. Thank you very much!
See also: Die l(i)ebenswerte Stadt – ADFC Stadtteilgruppe Treptow-Köpenick | Radfahren im grünsten Bezirk Berlins
1 Like
Hi Rob,
I’m back in Berlin now and will be testing a few things with my Lego NXT robot. I want to attach sample images of cars, motorcycles, bicycles, and people to the side of the NXT so that these images can be correctly recognized by the S2 camera as they pass by. Do you have any examples of the geometric dimensions (length/height/contours) of objects that can be detected by the S2 sensor?
We are currently preparing for the next event on September 20th/21st at the FEZ Berlin, where we’ll be busy building with Lego and counting with the S2 sensor.
1 Like
I don’t think there is such a standard. The AI has been ‘trained’ using hundreds of tagged images, not a set of parameters, so we can’t always know what it will recognise as one object or another.
It might be a case of testing it. It might also not always work with 2D images (although I’ve managed to get it to do this for certain images of a car, bike and pedestrian, but struggled with buses and trucks).
