First Road Tests

On 17. and 23. January I have carried out the first road tests of the software which goes along with this site: I mounted a video camera in my car and connected it to my capture software. This software then captured an image, based on the "difference" to the previous images. This algorithm simply compares pixel by pixel of two images and computes the sum of the absolute differences in each of the RGB channels. This sum is then divided by the number of bytes of each picture, resulting in a "normalised" value where 255 would be the maximum value.

It turns out that even when the image is not changing, there is an average noise change of 2 (out of 255), which means that the intensity of the captured light is fluctuating. This is to be expected, so a valu2 of 2 should not trigger yet any new image capture.

I did some very simple and only heuristic experiments: when the camera is looking from a static location at a scene which is basically not-changing, e.g. a group of buildings, then it takes in average about 5-10 minutes until the measured change of the value amounts to 10. This could warrant some more investigations into the statistics of the intensity change characteristics over the day. Naturally, the change of the sun's position will always lead to a change of the image pixel values over time, being dark at night and bright during the day, and also having colour shifts during morning and evening (more reddish).

When the camera is mounted in a moving vehicle, the changes are of course to be expected to me more drastically. There is, however, some constant pattern when driving on a road: the dark road surface ahead and the bright sky in the upper half provide some constancy in the overall image pattern.

When I experimented with the threshold of this normalised image change value, I found the value 40 to provide reasonable results: in cases of drastic changes, for example when turning into a new street, the image was updated frequently. But when driving steadily on a road, the image was not very often updated.

The setup used a shared internet connection through a Windows Mobile phone, so that the captured image could directly be uploaded to the server.

The web pages provide an overview on all captured pictures, as well as the image that was captured most recently.

Further plans: integration of GPS for position update, individual privacy setting for not giving public access to captured images, and multi-user logon to this site.