As new technologies like artificial intelligence begin to permeate the public sector, both in Australia and internationally, councils can transform the way they deliver services and keep communities safe.
According to a New South Wales government research report titled Automated Decision-Making in NSW, computer vision is being used across various government services, with notable projects including:
- Counting swimming pools and assessing rooftop solar panels
- Monitoring assets for maintenance needs
- Identifying issues in water management systems
- Enhancing public safety through improved monitoring
These diverse applications highlight the transformative potential of computer vision technology in modernising government operations and service delivery. In this context, the City of Greater Geelong is setting a new benchmark by leveraging advanced technology, data analytics and AI to drive innovation in pedestrian and road safety.
Traditional methods vs computer vision
Traditionally, assessing road safety at high-risk intersections involved manual surveys where data collectors would physically observe and record traffic and pedestrian behaviour. This method, while useful, is labour-intensive and provides only intermittent snapshots of traffic conditions. It also struggles with capturing data during off-peak hours or adverse weather conditions, limiting its effectiveness.
Computer vision offers a transformative approach. By utilising AI to analyse digital images and videos, computer vision systems can continuously monitor and assess traffic and pedestrian behaviour, significantly enhancing the ability to deliver a more accurate and complete understanding of local conditions.
How computer vision works
Computer vision technology for traffic and pedestrian movement involves several processes, including object detection, tracking and categorisation. The system starts with high-resolution cameras strategically placed at key locations to capture continuous video footage of the environment. Sophisticated algorithms trained on diverse datasets are employed to detect and classify various objects within the footage such as cars, heavy vehicles, bicycles and pedestrians. The system’s ability to accurately categorise these objects enables monitoring and analysis of traffic and pedestrian behaviour.
Once objects are detected and categorised, the system tracks their movement from frame to frame in real time, analysing their speed, direction and movement patterns. By analysing these trajectories, the system can identify potential hazards, such as a vehicle approaching a pedestrian crossing or a community member stepping into the crossing outside the designated area. The categorisation of objects allows detailed reports and insights, such as traffic density by vehicle type or pedestrian traffic patterns. This data is invaluable for optimising traffic management, enhancing safety measures and informing urban planning decisions to improve overall traffic flow and pedestrian safety.
Near misses can be identified by analysing instances where objects come extremely close to each other without resulting in a collision. Using spatial and temporal data, the system evaluates the proximity and movement patterns of objects to estimate the likelihood of potential accidents. Insights gained from these near misses enable the identification of high-risk scenarios, providing valuable data to refine traffic management strategies and mitigate potential hazards. This method supports detailed safety assessments and empowers data-driven decisions to enhance pedestrian safety and optimise overall traffic flow.
Case study: Ginn Street and Western Beach Road
The city deployed a sophisticated computer vision solution for traffic analysis and monitoring at the Ginn Street and Western Beach Road intersection on the Geelong Waterfront. This location was identified as a high-risk area due to anecdotal feedback reporting frequent near misses at the pedestrian crossing.
When using this technology, a crucial step is defining within the image frames the key capture points and areas, known as gates and zones. Virtual gates represent where data on pedestrian and vehicle movements will be collected including the type of object (e.g. car, motorcycle, pedestrian) and its trajectory.
Gate 1 (Ginn Street Entry) and Gate 3 (Ginn Street Exit) monitor positive direction traffic, recording 214 and 178 entries and exits respectively. Gates 6 and 7 monitor Western Beach Road traffic, with a significant number of entries and exits (4,825 and 3,855 respectively). Gate 5 monitors pedestrian traffic in both directions with 209 entries and exits. Zone 4 has a total of 9,062 entries and exits. Notably, there were no near-miss incidents recorded during this time period.
A video feed of the mobile unit, recorded between 10:00 and 11:00 on 16 August 2024 recorded 155 objects in the south direction, with a significant majority being cars. The north direction recorded 121 objects, also predominantly cars.
Near miss events at the Ginn Street pedestrian crossing were also recorded over a 40-hour period. Each near miss is logged, with unique identifiers assigned to individual incidents. The safety indicator values provide a quantifiable measure of risk, with lower values signalling more dangerous interactions between objects at the intersection. The precise timestamps offer insight into the timing and frequency of these events, enabling targeted interventions. By analysing this data, the city can better understand the dynamics at this intersection and implement strategies to enhance safety, such as traffic signals, improved lighting or redesigning pedestrian crossings.
Lessons learned
Computer vision holds immense potential for enhancing safety, reducing accidents and protecting lives. The city’s experience has provided valuable lessons for other local governments seeking to use computer vision for road safety and road-based applications.
Key takeaways include:
- Perform comprehensive site assessments to evaluate the intersection layout, analyse traffic patterns and determine the most effective sensor placement locations
- Ensure the AI is configured correctly to focus on capturing high pedestrian traffic areas and integrate insights into decision-making processes as early as possible
- Balance data costs by prioritising data feeds, while ensuring data is comprehensive and continuous for a complete traffic overview
- Utilise data insights to inform infrastructure changes, thereby improving pedestrian safety measures and enhancing overall safety at intersections
Geelong’s commitment to digital solutions
The city’s 2024-29 Organisation Strategy highlights its forward-looking commitment to leveraging digital and data foundations.
This approach extends beyond road and pedestrian safety. For instance, the city is embracing technology like rainfall sensors, enabling more effective responses to localised weather-related events and improving its ability to manage stormwater efficiently, reduce flood risks and strengthen overall infrastructure resilience.
Other examples include the use of high-fidelity 3D building and tree models, which allow for more simulations of the urban environment. These models are important for planning future growth and promoting environmental sustainability more effectively.
As the City of Greater Geelong continues to embrace the possibilities of digital and data-driven solutions, its commitment to innovation raises intriguing possibilities for the future. What if advanced technologies like computer vision could be further integrated to not only enhance pedestrian and road safety but also transform the way cities address challenges across all sectors? Geelong’s innovative mindset goes beyond solving today’s challenges – it’s embracing a ‘what if?’ perspective to build a forward-looking future where technology positively influences how Council leadership makes data-driven decisions, engages with communities and creates sustainable, inclusive urban spaces.
