Vehicle tracking and traffic monitoring at an intersection using an uncalibrated stereo vision system

Abstract

M.Ing. Traffic has become an extreme irritation and costly entity to deal with in recent years. Gone are the days where one could simply widen roadways to increase flow rates due to space constraints. Traffic costs countries billions of dollars per annum and thus the need to alleviate traffic congestion. Many technologies are currently available that can be used to lower the traffic density at an intersection, one of them being the use of cameras. Not only are digital cameras dropping in price, but the associated cost of maintenance is low. Distance information of a scene can thus be calculated via a visual system and from this information advanced control can be implemented in order to maximise traffic flow through an intersection. A traffic simulator was coded and analysed in order to validate the use of a visual system for increasing the amount of cars passing through the intersection per unit time over the current fixed timing system. Two different algorithms were compared to the current fixed timing scheme using a traffic simulator. The results showed that an improvement can be achieved over the current fixed timing scheme (of up to 19.92%). The use of stereovision as a method of attempting to monitor traffic flow is discussed. Vehicles were tracked using 13 trackers and the distance away from the stereo setup was calculated and compared to the actual distance away from the stereo setup. The best results found that with a baseline distance of 1500mm the average error in determining the distance of a vehicle was 16.46m. Although this error is quite large, it is still possible to monitor traffic flow using stereo vision with these inputs. Some of the issues that may cause these errors are camera quality, camera calibration and variable lighting conditions.

Original document

The different versions of the original document can be found in:

• http://hdl.handle.net/10210/5333