Estimating freight origin-destination activity using video and data

In this PhD project machine learning and video analytics solutions will be applied to automate the post-processing of captured data by developing methods that integrate multiple data sources for estimating freight origin-destination (OD) activities using video and traffic counts.

The estimation of freight movement activity in urban areas is an important consideration in freight transport planning, modelling and network congestion management. Developing new insights into freight journeys can enhance our understanding of travel patterns, network management, and identification of potential freight routes for safe and efficient deliveries.

Reliable OD information is important for strategic transport models that are used for future infrastructure investment and for evaluating the logistics consequences of economic development and city expansion.

The PhD approach will be focused on infrastructure-side solutions, including a network of low-cost cameras to be installed at strategic locations within the study area, supplemented by traffic counts collected from inductive loop detectors.

To demonstrate the feasibility of the approach, this PhD will undertake a field trial and PoC study on a confined network in Melbourne.

Participants

• Swinburne University of Technology

Project background

Despite decades of research on estimation of freight origin-destination (OD) activities, measurement and estimation of urban freight travel demand remains a big challenge. The key methodologies for freight movement determination includes roadside manual OD surveys (which are costly to conduct and require considerable human and financial resources), or using technologies such as cameras, automatic number plate recognition, Bluetooth scanners, weigh-in-motion sensors, inductive loop detectors as well as in-vehicle sensors, mobile phones and GPS for vehicle tracking.

All of these methods require significant post-processing of collected data and have been reported to have low estimation accuracy. However, out of all these methods, infrastructure-side solutions have been found to be the most reliable and cost-effective methods.

In this research we propose to apply machine learning and video analytics solutions to automate the post-processing of captured data by developing methods that integrate and fuse multiple data sources for estimating freight OD activities using video and urban sensing data such as traffic counts and other available information.

Project objectives

The project will have the following key objectives and tasks:

1. Establish current state of freight OD estimation using video analytics.
2. Identify and confirm a study area in Melbourne that can be used as a testbed for freight movement data collection and trialling of the proposed new methodologies.
3. Identify and install low-cost video camera solutions and traffic counting sensors to cover the study area’s entry and exit points and circulation areas and undertake data collection.
4. Develop a new framework for data integration, filtering, and fusion.
5. Develop new state-of-the-art video analytics methodologies and new approaches based on machine learning and deep neural networks techniques.
6. Develop a framework for the rollout of solution to cover the wider Melbourne Metropolitan region.
7. Disseminate the findings through annual workshops where the student would present an update to iMOVE and relevant industry key stakeholders and obtain feedback on progress to help the student improve research outcomes.

Profile of researcher, Shihan Xu

Find out more about this project’s researcher, including what questions the research will answer, the obstacles Shihan has overcome, what comes next after her PhD, and also hear from her supervisor, Professor Hussein Dia. Read the profile at: Shihan Xu – iMOVE PhD student.