No abstract

We

Abstract-Many vision-based automatic traffic monitoring systems require a calibrated camera for computing speeds and length-based classifications of tracked vehicles. A number of techniques, both manual and automatic, have been proposed for performing such calibration, but no study has yet focused upon evaluating the relative strengths of these different alternatives. We present a taxonomy for roadside camera calibration that not only encompasses the existing methods (VVW, VWH, and VWL) but also includes several novel ones as well (VVH, VVL, VLH, VVD, VWD, and VHD). We also introduce an overconstrained approach that takes into account all the available measurements, resulting in reduced error as well as overcoming the inherent ambiguity in the single-vanishing-point solutions. This important but oft-neglected ambiguity has not received the attention that it deserves; we analyze it and propose several ways of overcoming it. Our analysis includes the relative tradeoffs between twovanishing-point solutions, single-vanishing-point solutions, and solutions that require the distance to the road to be known. The various methods are compared using simulations and experiments with real images, showing that methods that use a known length generally outperform the others in terms of error, and that the overconstrained method reduces errors even further.

Vision-based automatic traffic monitoring systems require a calibrated camera in order to measure the speeds of tracked vehicles. Typically this calibration is done by hand. We present an automatic technique to calibrate the camera for typical viewpoints on highways using a realtime boosted cascade vehicle detector (BCVD). Image processing is used to estimate the two vanishing points, from which the camera height, focal length, and pan and tilt angles are calculated. The key contribution of the proposed approach is its applicability to a wide variety of environmental conditions. The technique does not rely upon background subtraction, nor does it require scene-specific features such as pavement markings. As a result, it is unaffected by the presence of shadows, adverse weather conditions, headlight reflections, lack of ambient light, or spillover caused by low-mounted cameras. Speed estimation within 10% of ground truth is shown for sequences obtained during daylight, nighttime, and rain, and including shadows and severe occlusion.

2Because of a recent federal initiative, states are now required (as of June 2008) to collect and 3 submit motorcycle VMT data to the FHWA. These data are needed to obtain better counts of 4 motorcycles to evaluate their impact on crashes and traffic flow. However, there is concern 5 about the quality of data submitted. Many states have identified problems with using automatic 6 traffic recorders to account for motorcycle traffic. Existing sensors exhibit difficulties in 7 counting motorcycles that travel side by side or close behind each other, they have difficulty in 8 distinguishing larger motorcycles from passenger vehicles, and magnetic counters in particular 9 do not sense motorcycles that do not pass over or travel close enough to the sensor.

A novel method is presented for automatically visually monitoring a highway when the camera is relatively low to the ground and on the side of the road. In such a case, occlusion and the perspective effects due to the heights of the vehicles cannot be ignored. Using a single camera, the system automatically detects and tracks feature points throughout the image sequence, estimates the 3D world coordinates of the points on the vehicles, and groups those points together in order to segment and track the individual vehicles. Experimental results on different highways demonstrate the ability of the system to segment and track vehicles even in the presence of severe occlusion and significant perspective changes. By handling perspective effects, the approach overcomes a limitation of commercially available machine vision-based traffic monitoring systems that are used in many intelligent transportation systems (ITS) applications. The researchers are targeting this system as a step toward a next generation ITS sensor for automated traffic analysis.

Decision support for real-time traffic management is a critical component for the success of intelligent transportation systems. Theoretically, microscopic simulation models can be used to evaluate traffic management strategies in real time before a course of action is recommended. However, the problem is that the strategies would have to be evaluated in real time; this might not be computationally feasible for large-scale networks and complex simulation models. To address this problem, two artificial intelligence (AI) paradigms—support vector regression (SVR) and case-based reasoning (CBR)—are presented as alternatives to the simulation models as a decision support tool. Specifically, prototype SVR and CBR decision support tools are developed and used to evaluate the likely impacts of implementing diversion strategies in response to incidents on a highway network in Anderson, South Carolina. The performances of the two prototypes are then evaluated by a comparison of their predictions of traffic conditions with those obtained from VISSIM, a microscopic simulation model. Although the prototype systems’ predictions were comparable to those obtained by simulation, their run times were only fractions of the time required by the simulation model. Moreover, SVR performance is superior to that of CBR for most cases considered. The study results provide motivation for consideration of the proposed AI paradigms as potential decision support tools for real-time transportation management applications.

A novel method is presented for automatically visually monitoring a highway when the camera is relatively low to the ground and on the side of the road. In such a case, occlusion and the perspective effects due to the heights of the vehicles cannot be ignored. Using a single camera, the system automatically detects and tracks feature points throughout the image sequence, estimates the 3D world coordinates of the points on the vehicles, and groups those points together in order to segment and track the individual vehicles. Experimental results on different highways demonstrate the ability of the system to segment and track vehicles even in the presence of severe occlusion and significant perspective changes. By handling perspective effects, the approach overcomes a limitation of commercially available machine vision-based traffic monitoring systems that are used in many intelligent transportation systems (ITS) applications. The researchers are targeting this system as a step toward a next generation ITS sensor for automated traffic analysis.