Automatic passenger counting with virtual sensors
Count people without personal data & complex hardware
Who needs passenger counting data and related information system?
Passenger counting data are useful for both passengers and public transport operators.
Especially after the COVID-19 pandemic, passengers in large cities express a strong demand to know “in real time” how many other passengers they can expect to travel at what time on which line. This demand for real-time information corresponds to several distinct needs of the users:
- First, people want to plan their trip in an optimal way, before leaving home, leaving the office etc. By optimum way, we understand: avoid waiting in queues, avoid large crowds of people in general, have a seat if possible, or at least, have enough space for standing comfortably. In large cities with multiple public transport lines, people want to know in advance, which combination of transport lines together with other options like walking a bit, self-service vehicles and so on will offer them the best compromise between total trip duration and travelling comfort. This requires predictive data about the number of passengers expected within the next hour, approximately, over an entire multi-modal transport network.
- During their trip, most of the passengers have their eyes on the smartphones: checking news, checking social media, but also checking if their planned trip is still the best option. If a bus or train is too crowded, does it make sense to wait for the next one? The kind of data required here is about the number of passengers currently riding the next 3 buses or trains of the same line.
- If waiting for the next bus is not a good option, then the question arises what would be the best alternative. This is especially true when there is an incident on one of the lines and suddenly the platforms of this line get overcrowded. People want to know what is the best alternative way to travel, from where they are right now, to their destination. That need is shared by the transport operator, who also wants to distribute the crowd of passengers in the most convenient manner for all of them.
In order to answer to the first question about predictive passenger numbers over a whole network, a public transport operator needs automatic passenger counting and a related information system, where extensive statistical ridership data of all their transport lines, over the year, over the days of the week and over the distance of a line can be managed and analyzed. They need to know, for instance, how many people usually take bus line X from stop A to stop D on a Monday at 11: a.m. The traditional manner to get such data was periodic manual passenger counting. This is however very time-consuming, and it offers only a limited statistical coverage – you cannot send somebody to count passengers every hour 24/7, on every line, during two weeks. Usually, only selected time slots like rush hours and one or two additional slots in between are evaluated.
Since many years, such manual counting data have been the foundation for optimization of public transport, by adjusting transport schedules to people’s needs:
- Add more buses at certain times and lines, and reduce the frequency at other times
- Add additional stops in areas with crowded boarding, and reduce stops in areas with few passengers
In the age of digitization, various solutions for automatic passenger counting have been developed.
What is an APC (Automatic passenger counting) system?
Automatic passenger counting systems, or APCS, allow counting the passengers going on board and leaving a public transport vehicle – like buses, tramways, trains – at each stop. Coupled with a telematics system for data transmission, an APC can provide real-time ridership information.
Such data is very useful for any public transport operator in order to optimize their service, as explained before. Real-time information on vehicle crowding are also more and more requested by the passengers themselves. Of course, the data are also used to measure the performance of a public transport operator and to report the performance to the local government.
Current automatic passenger counting market: Hardware-based automatic passenger counting technology
Most of today’s automatic passenger counting systems (APCS) are based on hardware sensors. The technology behind the sensor devices varies:
- In many Asian countries, RFID detectors are installed at the entries and exits. People put their mobile phone or RFID badge on the detector when entering and when leaving the line. They only pay for the actual section of the transport line they have used. On the other continents, this technology is still rather uncommon because passengers usually pay a flat rate per month or per day, and they do not “badge” when exiting.
- In Europe and in the USA, APC systems are typically based on optical devices using infra-red beams, 3D cameras or visual light beams placed at the entrances and exits.
- One alternative technology is based on measuring the axle load. The measurement data needs to be calibrated carefully in order to reflect the real passenger numbers: school children travelling in the morning have a different average weight per person than other passengers throughout the day. Baby strollers or suitcases have low weight, but require significant space. Even though passenger statistics can compensate for such bias, the complex vehicle dynamics on buses and the variation of driving styles make the calibration of axle load data for passenger counting purposes very difficult.
- A very recent technology is based on counting WIFI data. A majority of passengers are equipped with one (or more) smartphones that emit unique WIFI IDs even when the WIFI is “switched off” by the user. As satellite navigation offers limited coverage in cities due to tunnels and urban canyons, most smartphones use WIFI networks for localization in addition to GPS, even when the user “switches off” the WIFI for data exchange (unless they totally deactivate it via advanced setup). Every mobile phone emits unique traces which can be counted when entering and when leaving a bus. Together with RFID detectors, this is one of the technologies with the greatest accuracy, but it may also rise concerns about cybersecurity and the protection of personal data.
A low-cost alternative to an automatic counter is crowd-sourced data: people are asked to evaluate the perceived passenger load themselves on a mobile app or a web interface, from green (comfortable) to orange (somewhat crowded) to red (overcrowded). This obviously only works when a significant number of passengers is equipped with smartphones AND is willing to use the app, 24/7 and over the whole network. The feasibility of this option will greatly vary between cultures and countries. Alternatively, bus drivers can input such data themselves on an app. In short, APC systems based on crowd-sourced data are limited by subjective perception, lack of precision and the fact that they will not work out in every context.
In any case, the weak point of most high-end APCS systems is their complexity in regard to calibration, reliability and personal data protection.
Automated passenger counting with a software-based virtual mass sensor
COMPREDICT’s software-based technology uses already available vehicle data like speeds, torques and accelerations to calculate the loaded vehicle mass in real time. We do not need additional hardware devices except for a telematics system capable to log and transfer the vehicle data.
Together with a telematics company we have proven that our virtual sensor for loaded vehicle mass, which natively takes into account the vehicle dynamics, can be used as a robust passenger counting system on city buses.
Case study: Automatic passenger count data on a bus line
In this pilot project with one of our partners, we have used real-time GPS data, engine and vehicle speeds for mass calculation and, derived from the mass, as a passenger counter. On selected trips, we did some manual counting for comparison and were able to successfully validate our virtual sensor. At project start, we just needed some basic vehicle data to parameterize our algorithms.
We have shown in this project that we are able to estimate the number of people on buses with a precision of +/- 5% on urban buses.
Let’s discuss your use case!
Current status: COMPREDICT & telematics
To telematics providers, we offer an easy interface based on REST API and SDK so that we can perform the calculations on our algorithm hub AI-CORE without actually storing any customer data there. The algorithm hub is separated from the databases where the input and results data are stored. Both are hosted on the cloud. The cloud provider is selected together with every customer.
We are currently dealing with the first deployment of our solution and in discussion with other different providers of telematics solutions worldwide to make our solution available for fleet managers in the most convenient manner.
What’s behind a virtual sensor: how does our passenger counter work?
Our technology is based on AI and machine learning. We use already available vehicle data like engine torque and speed, wheel speed and lateral accelerations as input data. These data are collected with so-called data-loggers or telematics devices which are usually transferred to a (cloud-hosted) database from our customers.
The data from the database are used by our machine-learning algorithms on AI-CORE, our algorithm hub, via an API access or a request approach. The results can be provided to customers either as data stream, via REST API and SDK or, especially for pilot project, via our convenient and ready-to-use web-interface for visualization, AI-VIEW.
The vehicle mass itself is calculated by one of our proprietary algorithms, a virtual sensor for vehicle mass and vehicle overload detection.
+49 (0)6151 3844614
64283 Darmstadt, Germany