A pilot study on footprint posture classification of passengers in light rail public transport via deep convolutional neural networks

The public transportation industry is an essential mean of commuting within major cities and rural towns alike. Due to the high number of daily users, crowded environments can lead to limited seating requiring passengers to stand during their commute. Standing passengers are at risk during heavy braking scenarios and can also become a hazard to other seated passengers. Similarly, those who may not have a strong balance, or physically impaired such as the elderly endure a risk when vehicle braking is sudden. In this research, a method of analysing posture via simplistic pressure mat data is achieved. The integration of identifying a passenger's posture in relation to their balance may have the potential to inform public transport operators on the levels of braking that can be used based on their passengers at the time. Additionally, the identification of passengers using hand rail supports derived from floor mats only will inform operators of the potential risk during sudden braking. The results show an average classification accuracy of 99%, precision of 98%, recall of 98% and F-score performance of 98%.