Analysis of test data from an automatic vehicle monitoring (AVM) test

Data collected during a test of the location sub-system of four Automatic Vehicle Monitoring (AVM) systems in Philadelphia have been examined to determine characteristics useful in the employment of each technique in an operational system. A properly chosen algorithm can give accurate time point results of buses stopping at bus stops, but the addition of a door switch can provide greater reliability. Components of location error of each system can be fit graphically on normal probability paper, using linear segments with two different slopes. Some of the error components can be fit very well by a probability distribution composed of a sum of two normal distributions. The resultant distribution has more small errors and more large errors than would result from a single normal distribution. Although most systems exhibit a statistically significant bias, which could be removed during the location determination process, the amount is small compared to system requirements. Incorporation of signpost data improved LORAN-only accuracy approximately threefold.