DEVELOPMENT AND TESTING OF INCIDENT DETECTION ALGORITHMS, VOLUME 2: RESEARCH METHODOLOGY AND DETAILED RESULTS

Abstract

The development and testing of incident detection algorithms was based on Los Angeles and Minneapolis freeway surveillance data. Algorithms considered were based on times series and pattern recognition techniques. Attention was given to the effects of geometrics, sensor configuration and weather, and methods were developed for detecting malfunctioning sensors and for identifying the lane of an incident. The second volume describes the methodology used for evaluating and calibrating incident detection algorithms, the specific time series techniques examined (exponential smoothing and traffic correlation) and the decision tree structure for algorithms. Detailed results and discussions pertaining to all aspects of this study are presented here. /FHWA/