Video and image processing is important for traffic surveillance, analysis and monitoring of traffic conditions in many cities and urban areas. Present-day traffic management applications use video image processing to automatically analyze the scene of interest and extract information, such as speed capture, for traffic surveillance and control.
In particular, vehicle velocity measurement is one of the goals of traffic video surveillance. Vehicle velocity measurement may be an inherent part of a traffic video analyzer that recognizes and measures many other vehicle characteristics (licence plate, model, color, etc.) based on one or multiple cameras. There are three types of traffic video analyzers: trip line, closed-loop tracking, and data association tracking.
Trip line systems are based on user defined detection zones in the input frame from the camera. Video processing starts when a vehicle approaches one of those zones. The system estimates vehicle speed by measuring the time for a vehicle to move through a detection zone of known length.
Closed-loop tracking systems are based on full video frame processing and allow vehicle tracking continuously through the field of view of the camera. As soon as a vehicle is detected the speed is measured by the MacCarley algorithm. This algorithm performs a sequence of operations: background identification, vehicle detection, vehicle isolation from the background, flow front detection, handling shadows and artifacts and extracting a linear least square curve to measure the vehicle speed. The tracking system may also provide additional traffic flow data such as lane-to-lane vehicle movements.
Data association tracking systems identify and track a particular vehicle or groups of vehicles as they appear in the video frames. Image segmentation divides the image area into smaller regions (often composed of individual vehicles) where features can be better recognized. The feature extraction process examines the pixels in the regions for pre-identified characteristics belonging to vehicles. Artificial neural networks classify and identify vehicles, measure their traffic flow parameters, and even detect traffic accidents.
These areas are then tracked from frame-to-frame to produce tracking data for the selected vehicle or vehicle groups. The markers that identify the objects are based on gradients and morphology. Gradient markers utilize edges, while morphological markers utilize combinations of features and sizes that are recognized as belonging to known vehicles or groups of vehicles For the multi-camera case, the system uses data association tracking to gather travel time and origin-destination pair information by identifying and tracking vehicles as they pass from one camera video frame to another.