With the release of our Telraam S2 device we made a promise that in the not too distant future accurate traffic counts would become possible for motorised vehicles at night, in the dark too.
During recent months no specific night counting algorithm was running, but nevertheless - for testing purposes - we enabled the sensors to attempt counting overnight too. This is why you might have seen some counts on your device, or street segment even at night too. Now this initial testing period has ended it is clear that we need to develop a proprietary night counting method in order to achieve good night time accuracy.
Since the data of the test period is not accurate, we have decided to delete the inaccurate night time data from our database, to make sure that only accurate data is visible on telraam.net and via our APIs. This will also make sure that when accurate night time data is introduced, night time averages will not be influenced by the old inaccurate data. Daytime data is not affected in any way.
As - for the time being - we simply stop saving counts in our database which were recorded at night, you might see some discrepancy between daily counts on the screen of your Telraam S2 device, and the counts on the corresponding segment page, since the daily totals shown on the screen will include the inaccurate night time counts which are discarded by the Telraam database. This discrepancy will disappear when we deploy a new firmware containing the proprietary night counting algorithm. The development of this proprietary algorithm has started, and we are aiming at the coming winter for its deployment, so as the nights get longer again (in the Northern hemisphere), we will not lose sight of traffic anymore.
I’m curious if you plan to somehow mark or flag in the API that counts for some categories of vehicle are collected all day long, while others are only collected part of the day? Seems like for a naive API user that might be an easy to mistake and skew analysis.
This is still in development but there has been progress made by our team and the developers we work with, and we’re refining the algorithms. We still have work to do and testing as well, but we do hope to have it soon - though I can’t give you a date yet.
At night the devices are still active, but most of the time have trouble detecting anything they recognise because of the algorithm they use for detection. However, if there is some level of light, they may “detect” an object and it will appear on the screen. However, we have decided that until we update the algorithm, the reliability and accuracy of this data is not sufficient, so we ignore it on our reports.
The cut off for the count reports is linked to the quality / level of light that is reported during each update, so if this drops below a certain level, we will then ignore the data. This is done independently for each device, so the cut off will be different based on local conditions, particularly long / lat but also local weather, etc.
In theory, if you had stadium floodlights on your local street we would be able to count for 24 hours, … but you probably wouldn’t be able to sleep!
