For comparison, the US Department of Transport has an interesting page listing the unit costs of a number of congestion monitoring technologies. A set of inductive loop sensors at a single intersection cost 8.6-15.3k USD. Traffic CCTV installations (for the camera, electrical services, cabinet and installation) are estimated at 9.0-19.0k USD per camera, while a machine vision sensor at an intersection is estimated at 16.0-25.5k USD per installation. So we're looking at reducing the cost by two orders of magnitude.
The basic idea of our units is to house a battery, charging regulator and cameraphone in a steel box, as shown here:
The box is locked and mounted underneath a solar panel like this:
The charge in the phone is maintained by a 14W, 22V solar panel above the unit, and excess charge tops up a 7.2Ah battery pack via a charging regulator, in case of several consecutive overcast days. An arm extending from the solar panel allows the unit to be bolted to a wall or post, and the camera can be rotated through two axes.
A steel box is necessary for security but has the issue that it acts a Faraday cage, cutting out reception for the phone. In order to be able to upload images, we have to connect a wire from the phone's internal antennae to a wire coil outside the case.
|6:30am, not much activity|
|A little after 7am the morning rush is full swing|
|Evening rush at 6:15pm|
|Still slow traffic an hour later. It settles down by 9pm or so. The characteristics of the vision problem obviously change a lot at night, and we haven't looked into this a great deal yet.|
Given all this data, the focus is now on evaluating the accuracy of the software assessing the flow speeds on the road. We also have permission for installation of another two units at different exits of Makerere, so this will be interesting to be able to start looking at the correlation of traffic speeds at different places.