Passenger counting has been an essential part of the transit industry for decades. However, the cost of current automated passenger counter (APC) systems can be prohibitive for smaller fleets with small capital budgets and integration with existing bus controllers and agency databases is also challenging.
The objective of this IDEA project was to develop an inexpensive APC system that allowed simple integration with existing vehicle information systems to reduce cost. The system relates bus mass to passenger boarding and alighting using pressure transducers located in the air spring suspension system. A micro-controller installed on the bus conducted calculations at each stop event. A feed forward neural network and convolutional neural network algorithm were used to analyze the shape of the signal and predict boarding and alighting.
During the project, the system was integrated on four in-use transit buses around the Minneapolis and Saint Paul metropolitan area. The machine learning algorithms had 91% accuracy for a subset of the data used in training the algorithms. However, testing with the trained algorithms showed that the mass-based APC was 61% and 38% accurate in recording boarding and alighting events, respectively. Although the algorithms had poorer than industry standard accuracy for alighting events, especially, the system was able to estimate the mass of the bus at each stop accurately.
Future development work may look to use the mass-based system to augment existing infrared-based (IR-based) systems to improve their accuracy to new passenger counting standards. The team is extending the project findings to another research area that use the developed technology to better predict the driving range of all-electric transit buses by measuring instantaneous bus mass.
