Video-based incident detection

In the event of an incident on a roadway, it is important to be able to get to the scene as fast as possible, protect the incident site, and remove any victims and damaged vehicles, so that traffic can begin flowing again. The system described in this thesis detects incidents using real-time video. It separates the elements of an incident (using image processing methods) from the background images, extracts the features of these elements (such as number, length, width, and position), and then uses this information to determine the type of incident. The categories that this system can classify are objects on the side of road, an incident in a single lane, an incident in multiple lanes, and a traffic jam. From comparing the difference in weather detection regions of two kinds of background images, the system can also detect the changing of weather. The goal of this work was to develop a system with an improved Detection Rate (DR) and reduced False Alarm Rate (FAR).