A basic proof of concept application originally built as part of a job application to isolate and detect traffic. It turned out to be a successful project despite a lack of response from the target company. Unfortunately the source code for this was lost, but the video still exists.

This is one of the projects I would most like to revisit given the time, as with my improved experience as a developer I feel I could vastly improve its capabilities.

The advantages of this method are quite significant. Current automated systems mostly rely on object identification, which requires vast training and testing data, as well as having both false positives and false negatives. This system works on any static camera and ‘learns’ the road layout based on cars movement on a frame by frame basis. As more changes occur in any particular area a map of the hotspots is slowly developed. The system is designed to also remove areas which lack movement for extended periods of time and also adjust to the time of day by gradually fading out the hotspot in areas with no detected motion.

Once the estimated road layout is formed it begins to generate an image of the road without cars by only drawing parts of the road with no detected motion. This means it can compare an image of the empty road to the current video feed and subtract one from the other leaving only cars.

If I was to design this system again I would also incorporate estimated direction of travel on each detected road segment, which would also allow the system to detect and flag anything abnormal travelling in a different direction.