Efficiencies in replacement bus services during train outages

The Replacement bus patronage counting and wait time measurement project, conducted by iMOVE, the Department of Transport and Planning (Victoria), and Swinburne University of Technology, has been completed. Its final report is available for download below.

This project investigated two technologies for counting passengers boarding rail-replacement buses.

Background

The study focused on passenger waiting areas rather than on conditions inside buses. During scheduled track works and/or planned maintenance, buses replace train services. Myki card data and partial records from past disruptions are typically used to estimate travel demand to work out how many replacement buses are needed.

Additionally, customer service staff may have to determine patronage numbers on the spot. These inexact approaches mean alternative technologies that allow autonomous, reliable, accurate, and real-time measures of bus patronage are needed.

Methodology

The project team was keen to test and evaluate passenger counting technologies to automatically count patrons for bus-replaces-rail services. Important metrics included accuracy, strengths, weaknesses, feasibility, and the potential for other value-add options, such as estimating passenger wait time.

The team conducted field trials using two potential solutions, including video camera detection and pressure-sensing mats. The trials occurred at four Melbourne railway stations – Reservoir, Parliament, Pakenham, and Dandenong. Both trials were held on weekdays and collected in total more than 140 hours of data.

They comprehensively analysed the results to recommend a scalable and cost-efficient system.

Summary of results

The results showed both technologies offering high accuracy counts during expected working conditions. However, video analytics excelled for three of the use cases (passenger counting per bus, wait prediction, and bus run ID detection), with the sensor mat being most suitable for passenger counting.

Therefore, the project team recommended using only video analytics OR combining it with sensor mats.

Technology deployment considerations

Video analytics

For the first field trial, video scored 98.3 percent and 98.89 percent accuracy at Reservoir and Parliament, respectively; for the second field trial, it rated 72.91 percent and 95.31 percent for Pakenham and Dandenong, respectively. The cameras captured 25 frames per second.

The main advantages of video analytics included:

• High accuracy
• Versatility of positioning
• Supports per-bus counting, and
• Many other use cases.

However, trade-offs were the need for sufficient visibility of passengers, daily battery replacement and more regular lens cleaning.

Sensor mat

Meanwhile, in field trial 1, Reservoir achieved 80 percent and Parliament 84.45 percent accuracy; for field trial 2, Pakenham reached 87.25 percent and Dandenong 99.57% accuracy.

The advantages of using sensor mats were:

• Very high accuracy and consistency, and
• Primarily supported counting.

The main downside was that accurate data depended heavily on where the mat was placed. In Field Trial 1, the mat had to be positioned slightly away from the exact boarding point because buses did not always stop in the same spot.

While this offered some practical flexibility, it also meant that other station users walked across the mat, leading to inflated passenger counts. On-site barriers to guide passengers to step on the mat did not always succeed in the first trial, but were more effective in the second trial.

Both the video-based and sensor mat solutions will need to integrate with existing third-party services that the Victorian Government’s Level Crossing Removal Project (LXRP) uses. A cloud-based database will be developed to provide data access through standard APIs, so the solutions can communicate and easily share information with other systems.

These APIs can be hosted either on a cloud provider (such as Amazon) or on LXRP’s own servers. Final API design will form part of the next development stage once the relevant third-party services, their roles, and their data needs are confirmed.

Business considerations and implications

The video analytics solution uses established, commercially available off-the-shelf hardware supported by advanced machine-learning and software tools. Using off-the-shelf components keeps costs down for large-scale deployment and reduces the risk of failures or reliability issues. Overall, both solutions would still need some design and support modifications if chosen for a larger trial or real-time passenger counting.

With adequate resourcing, the video analytics technology can be deployed within six months; the sensor mat, within nine months.

At each site, the video analytics system takes up to 10 minutes to install, and its camera and electronics are expected to last around a year. Regular installation and removal do not significantly contribute to wear and tear. However, to work effectively, a video system needs adequate lighting or risks a 25% drop in passenger-count accuracy. Poor lighting also impacted the cameras’ ability to recognise Bus Run ID accurately, with mean scores ranging from 60% to 73%. Sensor positioning to capture the front of the bus and improved lighting would help.

Sensor mats require regular cleaning and battery changes and generally need replacing after about six months of steady use. Installation takes 10–15 minutes and repeatedly putting the mat in and taking it out could damage the wiring and increase wear.

Potential value-adding features of both technology options

Both technology options allow for feeding real-time edge-computed data to cloud-based services for data storage, further analysis, and dashboarding. Edge computing means no video data needs to be stored or transferred over the network, which is ideal for privacy concerns. Automated facial blurring could be integrated as an option to minimise privacy or security concerns.

Video analytics could work as or with CCTV cameras or purpose-installed video cameras in weatherproof housing. Customer service staff could check and adjust the video camera’s position using a tablet or smartphone-based app that has been prototyped. Ideally, this task should take no more than a minute per camera.

A potential use case for video analytics is to compute the time passengers wait for a replacement bus. In-queue wait times for two peak-hour trials at Reservoir showed the median was just eight seconds. The maximum wait time was three minutes and 20 seconds. The technology can help analyse how space is used how people move through an area, particularly the effective placement of barriers and signage to guide passengers.

In the longer term, video analytics could also be used to detect anti-social behaviour, and/or violence. They may also be useful to work out when specific stops require service and as patronage evolves over time, such as due to work-from-home patterns. Videos could also monitor passenger movement across a network and their transfers from/to public and private transport modes.

As for the sensor mat, in the short term, it holds promise for crowd direction. Over the longer term, it can operate day and night to provide valuable data for improving transport services and infrastructure, including use at train stations, bus and tram stops, and for anonymously monitoring crowd movements.

Expected project impacts

Real time, accurate patronage data is important for delivering reliable replacement bus services during disruptions. The technologies trialled in this project show clear promise in giving our operational teams the insights they need to make faster, better-informed decisions.

Paul Reichl, Manager, Research and Development Program, Victorian Infrastructure Delivery Authority (VIDA) Services

RA key outcome of this project was demonstrating that reliable, real time data on passenger demand during disruptions is achievable in practice. The findings provide a strong foundation for improving decision making around replacement bus services and highlight the value of close collaboration between government and university research teams.

Professor Chris McCarthy, Department of Computing Technologies, School of Science, Computing and Emerging Technologies, Swinburne University of Technology

Download the final report

Download your copy of the final report, Replacement Bus Patronage Counting and Wait Time Measurement, by clicking the button below.

This work was completed in June 2022.

DOWNLOAD THE REPORT