Developed at Lund University in Sweden, this database solution links tables describing identified conflicts to the recorded video making it simple, for example, to select conflicts of a certain type and to play short video sequences containing only those conflicts. The tool allows to manually fit pre-defined shapes of road users (car, cyclist, pedestrian, etc.) in the images and to extract their trajectories and speed profiles with a certain time resolution. Based on these data, it is possible to calculate the most common safety indicators (e.g. TTC, PET).
The software supports multi-camera recordings and includes a set of tools for video conversion and camera calibration.
RUBA—a watchdog software tool
RUBA (Road User Behaviour Analysis) was developed at Aalborg University in Denmark. The tool’s basic functional unit is a detector—an area of the image that is monitored constantly for activity. Several detector types are activated by presence, idling (long-term presence) or motion in a certain direction, and one detector recognises traffic light colour. Several detectors connected by a set of logical rules can be used at the same time. For example, it is possible to detect encounters (a car and a bicycle arriving simultaneously) or pedestrians walking on red.
The tool is most efficient when the frequency of expected events is low. Under favourable conditions, it allows removal of up to 90% of original footage that does not include relevant situations.
Traffic Intelligence project
This project at Polytechnique Montréal in Canada includes several tools for detecting, tracking and classifying road users, using a feature-based tracking algorithm for analysis of main outputs, trajectory data and road user interactions, as well as diagnosis of behaviour and safety. It has been applied to many case studies related, for example, to pedestrian behaviour and the safety of cycling facilities, highway entry and exit ramps and roundabouts. The technology has been used by several research teams and companies around the world.
While it includes tools for the most common tasks, it is best thought of as a software library for the user’s own scripts. As all the code is open source, researchers can contribute new functionalities and replicate research results, and wider adoption is encouraged.
STRUDL: Surveillance Tracking Using Deep Learning
STRUDL is an open-source and free framework for tracking road users in videos filmed by static surveillance cameras. It uses a deep learning object detector, camera calibration and tracking to create trajectories of e.g. road users, in world coordinates. It was designed to facilitate traffic safety analysis, using modern computer vision and deep learning. By creating trajectories in world coordinates, truly meaningful metrics and safety measures can be computed. STRUDL provides a Web UI that attempts to make it easy to use.
Using the program involves:
- Import videos and annotate images
- Train an object detector
- Provide camera calibration
- Perform tracking in world coordinates
- Download the tracks as csv files, and analyse them with whatever tools you like