Search Content

This report summarizes the research behind a real-time system for vehicle detection and classification in images of traffic obtained by a stationary CCD camera. The system models vehicles as rectangular bodies with appropriate dynamic behavior and processes images on three levels: raw image, blob, and vehicle. Correspondence is calculated between the processing levels as the vehicles move through the scene. This report also presents a new calibration algorithm for the camera. Implemented on a dual Pentium PC equipped with a Matrox Genesis C80 video processing board, the system performed detection and classification at a frame rate of 15 frames per second. Detection accuracy approached 95%, and classification of those detected vehicles neared 65%. The report includes an analysis of scenes from highway traffic to demonstrate this application.

This report covers the creation of a system for monitoring vehicles in highway on-ramp queues. The initial phase of the project attempted to use a blob tracking algorithm to perform the ramp monitoring. The current system uses optical flow information to create virtual features based on trends in the optical flow. These features are clustered to form vehicle objects. These objects update themselves based on their statistics and those of other features in the image. The system has difficulties tracking vehicles when they stop at ramp queues and when they significantly occlude each other. However, the system succeeds by detecting vehicles entering and exiting ramps and can record their motion statistics as they do so. Several experimental results from ramps in the Twin Cities are presented.

This report presents a real-time system for pedestrian tracking in sequences of grayscale images acquired by a stationary camera. Researchers also developed techniques for recognizing pedestrian's actions, such as running and walking, and integrated the system with a pedestrian control scheme at intersections. The proposed approach can be used to detect and track humans in a variety of applications. Furthermore, the proposed schemes also can be employed for the detection of several diverse traffic objects of interest, such as vehicles or bicycles. The system outputs the spatio-temporal coordinates of each pedestrian during the period the pedestrian is in the scene. The system processes at three levels: raw images, blobs and pedestrians. Experimental results based on indoor and outdoor scenes demonstrated the system's robustness under many difficult situations, such as partial or full occlusions of pedestrians. In particular, this report contains the results from a field test of the system conducted in November 1999.

The authors present a vision-based method for monitoring crowded urban scenes in an outdoor environment: background detection, visual noise from weather, objects that move in different directions, and conditions that change from day to evening. Several systems of visual detection have been proposed previously. This system captures speed and direction as well as position, velocity, acceleration, or deceleration, bounding box, and shape features. It measures movement of pixels within a scene and uses mathematical calculations to identify groups of points with similar movement characteristics. It is not limited by assumptions about the shape or size of objects, but identifies objects based on similarity of pixel motion. Algorithms are used to determine direction of crowd movement, crowd density, and mostly used areas. The speed of the software in calculating these variables depends on the quality of detection set in the first stage. Illustrations include video stills with measurement areas marked on day, evening, and indoor video sequences. The authors foresee that this system could be used for intersection control, collection of traffic data, and crowd control.

This report outlines a series of vision-based algorithms for data collection at traffic intersections. We have purposed an algorithm for obtaining sound spatial resolution, minimizing occlusions through an optimization-based camera-placement algorithm. A camera calibration algorithm, along with the camera calibration guided user interface tool, is introduced. Finally, a computationally simple data collection system using a multiple cue-based tracker is also presented. Extensive experimental analysis of the system was performed using three different traffic intersections. This report also presents solutions to the problem of reliable target detection and tracking in unconstrained outdoor environments as they pertain to vision-based data collection at traffic intersections.

Monitoring traffic intersections in real- time as well as predicting possible collisions is an important first step towards building an early collision warning system. We present the general vision methods used in a system addressing this problem and describe the practical adaptations necessary to achieve real-time performance. A novel method for three-dimensional vehicle size estimation is presented. We also describe a method for target localization in real-world coordinates, which allows for sequential incorporation of measurements from multiple cameras into a single target's state vector. Additionally, a fast implementation of a false-positive reduction method for the foreground pixel masks is developed. Finally, a low-overhead collision prediction algorithm using the time-as-axis paradigm is presented.

This report describes a real-time system for tracking pedestrians in sequences of grayscale images acquired by a stationary camera. The system outputs the spatio-temporal coordinates of each pedestrian during the period when the pedestrian is visible. Implemented on a Datacube MaxVideo 20 equipped with a Datacube Max 860, the system achieved a peak performance of over 30 framers per second. Experimental results based on indoor and outdoor scenes have shown that the system is robust under many difficult traffic situations. The system uses the "figure/ground" framework to accomplish the goal of pedestrian detection. The detection phase outputs the tracked blobs (regions), which in turn pass to the final level, the pedestrian level. The pedestrian level deals with pedestrian models and depends on the tracked blobs as the only source of input. By doing this, researchers avoid trying to infer information about pedestrians directly from raw images, a process that is highly sensitive to noise. The pedestrian level makes use of Kalman filtering to predict and estimate pedestrian attributes. The filtered attributes constitute the output of this level, which is the output of the system. This system was designed to be robust to high levels of noise and particularly to deal with difficult situations, such as partial or full occlusions of pedestrians. The report compares vision sensors with other types of possible sensors for the pedestrian control task and evaluates the use of active deformable models as an effective pedestrian tracking module.

This report presents a real-time system for pedestrian tracking in sequences of grayscale images acquired by a stationary CCD (charged-coupled devices) camera. The research objective involves integrating this system with a traffic control application, such as a pedestrian control scheme at intersections. The system outputs the spatiotemporal coordinates of each pedestrian during the period the pedestrian remains in the scene. The system processes at three levels: raw images, blobs, and pedestrians. It models blob tracking as a graph optimization problem and pedestrians as rectangular patches with a certain dynamic behavior. Kalman filtering is used to estimate pedestrian parameters. The system was implemented on a Datacube MaxVideo 20 equipped with a Datacube Max860 and on a Pentiumbased PC. The system achieved a peak performance of more than 20 frames per second. Experimental results based on indoor and outdoor scenes demonstrated the system's robustness under many difficult situations such as partial or full occlusions of pedestrians

This report evaluates the Minnesota Department of Transportation's (Mn/DOT) Salt Solutions program over the past two years. The evaluation documents the components of the program, describes the technology, and provides a detailed cost-benefit analysis. Recognizing the potential to reduce the level of salt and sand use, the maintenance division began a reduction initiative in District 1 during the 1996-97 snow and ice season. The Salt Solutions program sought to develop a set of tools and a system that allowed operators to make better application rate decisions, support those tools and systems with ongoing training, develop controls and measurements to track the effectiveness of the tools and training, and recognize improved performance. The program expanded statewide in the 1997-98 winter season. Results of this evaluation show that the program is cost-effective means of reducing the amount of salt and sand applied to Minnesota roadways while still maintaining a safe operating environment. In its first year, the program saved an estimated $177,000.

We propose methods to distinguish between moving cast shadows and moving foreground objects in video sequences Shadow detection is an important part of any surveillance system as it makes object shape recovery possible, as well a improves accuracy of other statistics collection systems. As most such systems assume video frames without shadows, shadows must be dealt with beforehand. We propose a multi-level shadow identification scheme that is generally applicable without restrictions on the number of light sources, illumination conditions, surface orientations, and object sizes. In the first level, we use a background segmentation technique to identify foreground regions that include moving shadows. In the second step, pixel-based decisions are made by comparing the current frame with the background model to distinguish between shadows and actual foreground. In the third step, this result improved using blob-level reasoning that works on geometric constraints of identified shadow and foreground blobs. Results on various sequences under different illumination conditions show the success of the proposed approach. Second, we propose methods for physical placement of cameras in a site so as to make the most of the number of cameras available.