The Transport & Planning department has developed software to process series of traffic images into vehicle trajectories.
The entire process of acquiring images up to analysis is described below:
- Image acquisition. This can be done with a camera from a helicopter, an Unmanned Aerial Vehicle (UAV), a balloon, or a high-rise building, etc.
- Image export. Conversion of the images from the acquisition system to a series of separate JPEG or (preferably) PNG images. The images must be named in a way that alphabetical order equals chronological order. We recommend putting the time of acquisition in the file name.
- Lens correction, stabilization (registration), orthorectification and cropping. This is done with our ImageTracker software. This program writes a series of images that cover the user-specified area and the scale factor from image to world should be the same everywhere in the image. It is possible that the area covered is not the same in all source images. In fact, the observation platform may fly along with traffic to a limited extent. Pixels that were not observed in an image are shown in 100% Red in the stabilized images.
- Moving object detection. This is done with our ObjectFinder software. This software determines the background of the scene in the stabilized images and finds objects (of a minimum size) that stand out from the background. There will always be objects (cars) that are indistinguishable from the background, but ObjectFinder usually finds most moving objects. The output of this software is a set of text files (one per image) describing the locations of the objects.
- Connect detected objects into trajectories. This is done with our CombineBlobs software. This software reads the lists of objects (found by ObjectFinder) and tries to link the objects into trajectories. Some false positives of the moving object detection will be filtered out in this phase.
- Analyze trajectories. This must be done by software (or manually) by the traffic researcher. We expect that software for this task will be developed by our PhD students in the coming months.
The use of the image processing software is not restricted to vehicle traffic research. It should be possible to track pedestrians, bicycles, bacteria, etc. It does help if moving objects keep a minimum separation. The presence of shadows can be confusing for the ObjectFinder software.