VaxCounting: Counting and Measuring Vehicles Speed

[Neochange86]

Thanks, It's so interesting, but I think It is so difficult to earn money with that. You need calibrate every camera separately and road conditions (inclination, etc) are too important.

Regards

[vaxtorsystems]

Hi, we have external calibration process using a vehicle driving under controlled speed via GPS. The internal calibration process is our camera self-calibration software which quite sophisticated. We use the same software in 3D reconstruction from video (see vaxtor.com) to model the camera motion in a 3D environment... so the whole calibration process takes about 1 hour, less time than mounting the camera on top of the post. The software does not work well on hard inclination and curves

[sticky170]

Not realy accurate is it?

[vaxtorsystems]

Hello... well it depends on the calibration process and camera reliability... for a "not bad" 700 USD camera we get 90% accuracy with no problem... during the day. For high speed shutter camera we can stay above 90% most of the time

[necromncr]

Look on the right size - the blue sign! It's the.. the..!!

[vaxtorsystems]

jejeje, the blue sign means you can make a "U" turn next exit in the highway :)

[Neochange86]

Hello, this demo is great, but how do you calculate depth without stereoscopic cameras? It's just curiosity, I am thinking about a pixel represents a different lenght at the bottom and at the top of the image.

Thanks

[vaxtorsystems]

Hello, the speed calculation is based on a function (analytic) extracted from the motion forecast and other parameters we get from out self-calibration software we developed for cameras. We also need a first vehicle driving at different (known) speeds for additional calibration purposes. In this case, the speed equation v = s/t assuming 's' as a distance based on pixel size, does not work, I can tell you that for sure.

[Shaunakde]

What happens in heavy traffic? Does it still recognize the vehicles if two blobs overlap?

[vaxtorsystems]

Hi... not just blobs anaylsis... we use SURF to detect local features in the images, shadow removal technics and Kalman filters to forecast motion and calculate the speed more accurately. Hence, the software works fine under dense traffic scenarios but it may show up lack of accuracy under heavy traffic jams counting from low angle camera.