Darcy Bullock

A neural network for image-based vehicle detection

Abstract Vehicle detection on roadways is useful for a variety of traffic engineering applications from intersection signal control to transportation planning. Traditional detection methods have relied on mechanical or electrical devices placed on top of, or embedded in, pavements. These systems are relatively expensive to install, tend to be unreliable over time, and are limited in their capabilities. Considerable research has been conducted in the area of machine vision for Wide Area Vehicle Detections Systems (WADS). These systems have typically employed conventional image processing and pattern matching algorithms, and many installations have been sensitive to varying lighting conditions, camera perspective, and shadows. In addition, these systems have often required large amounts of computing resources. This paper reports on the development of a new image based vehicle detection system that is based on a simple back propagation/feedforward neural network for tracking vehicles. Application of this concept in a field system is discussed and preliminary results are presented. These results suggest that the neural network vehicle tracking model can be used to reliably detect vehicles. In addition, the training capability of the neural network detection model permits the system to adapt to variations in lighting and camera placement. This should lead to simplified installation and maintenance of WADS.

Roadway traffic control software

Roadway traffic control is an important practical example of real time control applications, but effective control is hampered by a variety of organizational, financial and technical considerations. One major hurdle is the current reliance on outmoded field hardware and software. A systematic approach to traffic engineering software development could provide significant advantages with regard to software capability, flexibility and maintenance. Improved traffic controllers will likely be essential for many of the proposed intelligent vehicle highway systems (IVHS) applications. This paper describes the roadway traffic control problem generally and introduces a computable language that can be used for constructing real time traffic control software. This computable language is designed to be configured by a graphical user interface that does not require extensive software engineering training to use, yet provides much more flexibility and capability than possible by simply changing program parameters. The model is based upon the function block metaphor commonly used for constructing robust and efficient real time industrial control systems. The software model has been implemented in C on an open architecture traffic controller (OATC) hardware platform and demonstrated under simulated conditions for applications such as signalized intersection control, ramp metering, and communications with existing traffic control devices. System users can construct applications from a library of function blocks. The paper describes a demonstration application to freeway ramp metering control in Sacramento, CA. >

Analysis of feedforward-backpropagation neural networks used in vehicle detection

Abstract Current vision-based vehicle detection systems use image-processing algorithms to monitor the presence of vehicles on the roads. Recent research has shown that an artificial feedforward neural network can be trained to provide similar capabilities. A properly trained and configured network should be able to recognize the presence of vehicles in the images it has never been exposed to. This paper discusses the development of a feedforward-backpropagation neural network-based vehicle detection system that recognizes and tracks vehicles with satisfactory reliability and efficiency. Various issues that are important in selecting the optimal neural network model—like the architecture of the network including the number of hidden layers, their units, learning rule, tiling characteristics of the input image and the output representation of the network—are addressed in this paper. This paper also analyzes how the neural network internally learns the mapping knowledge of the input-output training pairs. The final section describes an output post processor that produces the traditional pulse and presence signals.

A neural network based vehicle detection and tracking system

Recent research has shown that feedforward neural networks can be trained to monitor vehicles on the roads (D. Bullock et al., 1993). A properly trained network should be able to recognize vehicles in the images it has never been exposed to. The paper discusses the development of such a neural network based detection and tracking model. The detection and tracking model was constructed on a PC using video tapes of traffic. A hybrid system architecture was developed to provide the necessary interface between the software and hardware modules. Two types of neural networks were investigated: standard feedforward networks and radial basis function (RBF) networks. Various tests were conducted to determine the optimal network model. The RBF network performed better than the conventional feedforward model. A success rate of 93% was achieved with the RBF network based detector model.<<ETX>>

IMPLEMENTATION VISION FOR DISTRIBUTED CONTROL OF TRAFFIC SIGNAL SUBSYSTEMS

The developments that have led to the construction of the 2070 controller are reviewed. The intelligent transportation system community has proposed many features and user services that will likely use this new controller. In general, many of the functions proposed for this controller, such as emergency vehicle preemption, transit priority, weather monitoring, dynamic lane assignment, enhanced malfunction diagnostics, and adaptive algorithms, are all technically feasible. To achieve widespread deployment of systems that integrate several advanced traffic management system features, however, a systematic method for integrating a variety of distributed computing subsystems must be thoughtfully defined. The fundamental benefits of adopting a distributed control model for traffic signal subsystems are described and summarized.

Implementation of a distributed control system for dynamic lane assignment

This paper reviews the basic signing technologies available to the traffic engineer for dynamically changing lane assignments. An example installation of dynamic lane assignment in Houston, TX is reviewed and referenced as the motivation for developing a simple traffic signal cabinet interface for dynamic lane assignment signs. Various candidate control architectures are summarized and a distributed control model is recommended due to the simple cabinet retrofit and the inherent malfunction management capabilities. The paper concludes by describing the distributed computing design comprising the sign control interface (SCI) residing in the cabinet and the sign control modules (SCM) residing in each sign.