Nancy L. Nihan

RECURSIVE ESTIMATION OF ORIGIN-DESTINATION MATRICES FROM INPUT OUTPUT COUNTS

The application of recursive prediction error techniques to the problem of estimating origin-destination patterns from input and output volume counts is described. Each algorithm deals with the special case where route choice between origin and destination can be ignored. A gradient algorithm developed by Cremer and Keller (1983) turns out to be a special case of a family of methods described by Ljung and Soderstrom (1983). After describing how the methods developed in Ljung and Soderstrom (1983) could be modified so that the resulting estimates satisfy natural constraints, a number of algorithm possibilities are tested. Generally, those algorithms employing Gauss-Newton search directions appear superior to gradient-based methods, while the constraining procedures improve accuracy.

ESTIMATING THE RISK OF COLLISIONS BETWEEN BICYCLES AND MOTOR VEHICLES AT SIGNALIZED INTERSECTIONS

Collisions between bicycles and motor vehicles have caused severe life and property losses in many countries. The majority of bicycle-motor vehicle (BMV) accidents occur at intersections. In order to reduce the number of BMV accidents at intersections, a substantial understanding of the causal factors for the collisions is required. In this study, intersection BMV accidents were classified into three types based on the movements of the involved motor vehicles and bicycles. The three BMV accident classifications were through motor vehicle related collisions, left-turn motor vehicle related collisions, and right-turn motor vehicle related collisions. A methodology for estimating these BMV accident risks was developed based on probability theory. A significant difference between this proposed methodology and most current approaches is that the proposed approach explicitly relates the risk of each specific BMV accident type to its related flows. The methodology was demonstrated using a 4-year (1992-1995) data set collected from 115 signalized intersections in the Tokyo Metropolitan area. This data set contains BMV accident data, bicycle flow data, motor vehicle flow data, traffic control data, and geometric data for each intersection approach. For each BMV risk model, an independent explanatory variable set was chosen according to the characteristics of the accident type. Three negative binomial regression models (one corresponding to each BMV accident type) were estimated using the maximum likelihood method. The coefficient value and its significance level were estimated for each selected variable. The negative binomial dispersion parameters for all the three models were significant at 0.01 levels. This supported the choice of the negative binomial regression over the Poisson regression for the quantitative analyses in this study.

Application of Prediction-Error Minimization and Maximum Likelihood to Estimate Intersection O-D Matrices from Traffic Counts

The use of prediction error and maximum likelihood techniques to estimate intersection turning and through movement probabilities from entering and exiting counts is considered here. A maximum likelihood estimator for situations when full information on turning movement counts is available is derived and used as a component for a maximum likelihood algorithm which only requires entering and exiting counts. Several algorithms based on minimizing the error between observed and predicted exiting counts are also developed. Some actual traffic data are collected and used to develop realistic simulations for evaluating the various estimators. Generally, the maximum likelihood algorithm produced biased but more efficient estimates, while prediction error minimization approaches produced unbiased but less efficient estimates. Constraining the recursive version of the ordinary least-squares estimator to satisfy natural constraints did not affect its long run convergence properties.

USE OF THE BOX AND JENKINS TIME SERIES TECHNIQUE IN TRAFFIC FORECASTING

This paper explores the use of recently developed time series techniques for short term traffic volume forecasts. A data set containing monthly volumes on a freeway segment for the years 1968 through 1976 is used to fit a time series model. The resulting model is used to forecast volumes for the year 1977. The forecast volumes are then compared to actual volumes in 1977. The results of this study indicate that time series techniques can be used to develop highly accurate and inexpensive short term forecasts. A discussion of the ways in which such models can be used to evaluate the effects of policy changes or other outside impacts is included.

Extracting Roadway Background Image: Mode-Based Approach

Traffic monitoring cameras are widely installed on streets and freeways in U.S. metropolitan areas. Video images captured from these video cameras can be used to extract many valuable traffic parameters through video image processing. A popular way to capture traffic data is to compare the current traffic images with the background image, which contains no vehicles or other moving objects, just background such as pavement. Once the moving vehicle images are separated from the background image, measurements of their number, speed, and so on can be obtained. Typically, background images are extracted from a video stream through image processing because it may be hard to find a frame without any vehicles for normal traffic streams on urban streets. This paper introduces a new method that can quickly extract the background image from traffic video streams for both freeways and intersections in a variety of prevailing traffic conditions. This method has been tested with field data, and the results are promising.

Freeway Traffic Speed Estimation with Single-Loop Outputs

Traffic speed is one of the most important indicators for traffic control and management. Unfortunately, speed cannot be measured directly from single inductance loops, the most commonly used detectors. To calculate space-mean speed, a constant, g, is often adopted to convert lane occupancy to traffic density. However, as illustrated by data from the present study, such a formula consistently underestimates speed whenever a significant number of trucks or other longer vehicles are present. This is because g is actually not a constant but, rather, a function of vehicle length. To calculate the value of g suitably, one needs to know the percentage of long vehicles or the mean vehicle length in real time. However, such information is not directly available from single-loop outputs. It is shown how the occupancy variance obtained from single-loop data can be used to estimate the percentage of long vehicles and how a log-linear regression model for mean vehicle length estimation based only on single-loop outputs can be developed. The estimated mean vehicle length is used to calculate the corresponding g-value in real-time to estimate speed more accurately. The speed estimations with corrected g-values are very close to the speeds observed by the speed trap in the present study.

Large population approximations of a general stochastic traffic assignment model

Recent interest in stochastic traffic assignment models has been motivated by a need to determine the stationary probability distribution of a networks traffic volumes and by the possibility of using time-series of traffic counts to fit and test travel demand models. Because of the way traffic volumes are generated as the sum of path flows from different origin-destination pairs, and because of the nonlinear nature of the process relating traffic conditions to traveler route selection, most plausible assignment models tend to be intractable. In this paper, we first pose a general stochastic assignment model that includes as special cases most models which have appeared in the literature, and then verify that the probability distributions of an equivalent Markovian model converge to a stationary distribution. We next verify that as the number of individual travelers becomes large, the general model can be approximated by the sum of a nonlinear deterministic process and a time-varying linear Gaussian process. The stationary distribution of this approximation is readily characterized, and the approximation also provides a means for employing linear system methods to estimate model parameters from a set of observed traffic counts. For the case where the route choice probabilities are given by the multinomial logit function, computationally feasible procedures for implementing the approximate model exist.

Identification and Correction of Dual-Loop Sensitivity Problems

Freeway traffic speed and bin volumes for different vehicle categories are typically collected by dual-loop detectors. Good quality dual-loop detector data are crucial for effective real-time traffic management systems and traveler information systems. However, loop detectors are subject to various malfunctions that can result in erroneous measurements. Previous studies indicated that loop sensitivity-level discrepancies between two single loops forming a dual-loop detector and unsuitable sensitivity levels of the single loops are two major causes of quality degradation in dual-loop data. This paper presents an algorithm and its implementation for identifying and correcting such loop sensitivity problems. The algorithm identifies dual-loop sensitivity problems using individual vehicle data extracted from loop event data and corrects dual-loop sensitivities through a two-step procedure: (a) remove the sensitivity discrepancy between the two single loops and (b) adjust their sensitivities to the appropriate...

Investigation into Shadow Removal from Traffic Images

Traffic surveillance cameras are becoming a viable replacement for inductive loop detectors. The effectiveness of these cameras, however, depends on video image processing algorithms that can alleviate common problems such as shadows, vehicle occlusion, reflection, and camera shake. Shadows have proved to be a major source of error in the detection and classification of vehicles. Three algorithms of increasing complexity are proposed to address the shadow problem. The algorithms each address the need to remove cast shadows from vehicles while preserving self-shadows, or those areas of a vehicle that are hidden from illumination. They are also geared toward real-time analysis, which requires that they can be implemented efficiently and cannot have complex training or learning requirements. The dual-pass Otsu method of shadow removal was the simplest in application but had the poorest performance. The proposed region growing technique, though showing considerable promise, failed when the pixel intensity varied widely in the shadow region. The last technique used edge imaging to recognize shadows as areas with few edges or with edges substantially similar to the background. This method clearly outperformed the other methods and was subsequently proved in a separate paper describing a prototype vehicle detection and classification system.

Quantitative evaluation of GPS performance under forest canopies

There is an increasing demand for use of the Global Positioning System (GPS) to navigate or track objects in the forest. However, objects near a GPS receiver antenna, such as tree leaves and branches, can reflect GPS signals and result in large position errors. Canopies in the forest will also block satellite signals and cause the GPS receiver to stop updating data. This is of practical significance for evaluating the performance of GPS in the forest environment. A field test was conducted to understand how large the position errors are and how long the position updates may be deferred under different levels of canopy densities. A digital camera was used to record the canopies over the test site. Image processing techniques, especially Otsus algorithm, were used and the canopy density was classified into three levels. The ANOVA was used to analyze the effect of canopy density on the GPS position errors. The result shows that the GPS position errors are significantly different under different canopy density levels. The GPS data-update frequency was also analyzed, and the result indicates that the scheduled position update intervals are lengthened due to the existence of forest canopies.