Project Overview
In April 2020, the Point-to-Point Transport Commissioner (the Commission) partnered with Cisco Systems Australia (Cisco) to complete a three-month trial using AI-enabled CCTV cameras to gain insights into taxi rank usage at Sydney’s Central Station. Use of taxi ranks is governed by the Point to Point Transport legislation, which provides rules for how and when taxi ranks should be used.
The Commission was aware that many taxis and non-taxis were not complying with rules around rank usage. Incorrect use of taxi ranks poses safety risks for drivers and pedestrians and impacts customer access to taxi services due to overcrowding.
Two AI-enabled CCTV cameras were installed at the taxi rank outside the Grand Concourse at Central Station to monitor and analyse activity on the rank and provide the Commission with images and information on what types of vehicles are using the rank and how they are using it. The AI in the CCTV cameras assesses the captured footage against specific criteria onsite and delivers this information to a CCTV Analytics Dashboard managed by the Compliance branch of the Commission.
The proof of concept was successful at demonstrating the opportunity for AI-enabled CCTV cameras to capture and analyse taxi rank activity to enhance compliance activities and increase customer safety. As a result, the Commission has continued the use of these cameras at Central Station and has allocated funding to extend this pilot to more taxi ranks across NSW.
New Opportunities
The Commission identified that the information gathered by AI-enabled CCTV cameras can be used to support Transport for NSW’s (TfNSW) system planning to provide a better experience for point to point transport users. These opportunities are being explored further during the extended pilot. In particular, the CCTV cameras gather useful information which can be used by TfNSW to provide a better experience for point to point transport users by:
- Measuring patronage including approximate people counts and passenger flow at a specific rank to inform planning needs
- Measuring rank usage patterns including peak times for taxis and passengers, to better inform service demand decisions for drivers and passengers
- Identifying service gaps for passengers who are waiting at ranks where there are no available taxi services
- Analysing rank usage by wheelchair accessible taxis and identify service gaps for these taxis.
Deployment of AI
The Commission is a regulatory body which relies upon a small number of compliance officers to gather insights on how taxi ranks are used and where compliance action is best targeted. In recent years, TfNSW has embraced the opportunities presented by AI to improve system planning and safety outcomes of the NSW Transport network and had already undertaken a number of AI enabled CCTV trials for other purposes prior to this project. Due to this, the Point to Point Transport Commissioner has a strong understanding of the benefits of AI and proposed that this technology could be leveraged to gather insights on the point to point transport system to inform compliance activity and improve system planning.
The Commission collaborated with Cisco, one of their industry partners, to co-design a proof of concept solution that would fit the Commission’s requirements. Following the success of the proof of concept, the Commission is now working with Cisco and two vendors to expand the trial to the additional locations and continue the innovation process and share learnings.
Compliance with AI Ethics Policy
The AI Ethics Policy provides a set of key principles that guide the ethical use of AI by the NSW Government and ensure best practice. These focus on community benefit, fairness, privacy and security, transparency and accountability.
The initial three-month trial phase occurred before the development of the AI Strategy and Ethics Policy. Despite this, the Commission was aware of the risks involved with the use of AI in this area and implemented appropriate governance and controls to ensure that the proof of concept had an effective outcome. The Commission implemented a Governance Officer to oversee the project and ensure alignment and compliance with relevant NSW Government legislation, policies and guidelines and the NSW Government AI Strategy and Ethics policy when it was implemented.
Assessment against AI Ethics Policy
Community Benefit
The proof of concept phase was limited to testing the feasibility of using AI enabled CCTV cameras to generate useful data insights. However, the expanded trial is expected to have significant customer benefits by enhancing the Commission’s existing compliance activity, increasing confidence in the safety of taxi ranks, and informing customer choice by increasing information available to customers.
Fairness
Issues around incorrect or unconfirmable categorisations by the AI resulting from poor image quality are being managed in the trial phase. The AI capability of the cameras is constantly evolving and improving its categorisations. The model is also constantly monitored and regularly tested to ensure that it continues to improve before future opportunities for the technology are investigated. The Commission considered the impacts of this technology on minority groups when developing the proof of concept and expanded trial. As a result, the Commission is planning to use the AI-enabled CCTV cameras to identify wheelchair accessible taxis and analyse their rank usage.
Privacy and Security
Under Road Transport Legislation, TfNSW has the legal authority to install and maintain cameras for the purposes of traffic control. As an enforcement body, the Commission is able to use these cameras to protect public safety and monitor compliance with the point to point transport legislation.
All data is held in accordance with the Privacy and Personal Information Protection Act 1998 (PPIP Act). All captured footage and associated data is protected with end-to-end encryption and stored securely in the cloud. Footage collected by the cameras is destroyed when it was no longer required. As images of people are collected by the AI, the software blurs out faces to protect identity.
Transparency
Signage was erected at the camera locations to inform the public of the presence of CCTV cameras and community feedback was sought from local residents. Trial locations were also made available on the Commission’s website which is regularly updated to include the locations of the cameras, the purposes for which the information is being collected, the intended recipients of the information, and the existence of any right of access to the information. The Commission have communicated directly with industry to inform point to point transport companies of the trials, and residents of the surrounding areas were consulted prior to the installation of the cameras.
Accountability
Data and analysis provided by the AI to the Commission through the dashboard are designed to enhance the compliance activities and decisions of the Commission. The AI does not make any decisions and accountability for decisions made using the dashboard remain with an authorised officer at the Commission. These officers are all trained to identify breaches and record evidence of non-compliance, and are aware of the limitations of the AI technology and the insights they receive.
Outcomes
Data and analysis provided by the AI to the Commission through the dashboard are designed to enhance the compliance activities and As a result of the success of the trial, the Commission has continued the use of the two original CCTV cameras at Central Station and has allocated funding to extend the pilot to up to 100 taxi rank locations across NSW for 12 months. The Commission expects the expanded trial to enhance the Commission’s existing compliance activity, increase public confidence in the safety of taxi ranks by deterring unsafe behaviours at ranks and informing customer choice about the availability of taxis.
Capabilities and skills required for the application of AI
The Commission and TfNSW used a combination of internal capabilities and skills and also leveraged industry knowledge and expertise to build the AI tool and implement this project. This collaborative approach allowed the Commission to access the best in business, share and test ideas and upskill their teams. Specifically, the Commission developed the new idea, the method and the process whilst industry experts filled any technical skill gaps.
Teams from both the NSW public sector and Industry came together for this project. The project team included team members from TfNSW and the Commission, CISCO, who is a worldwide leader in IT, and other small to medium vendors.
Technical skills
Data literacy, Artificial Intelligence, Machine Learning and Coding skills
The project team engaged Cisco and partnered with additional small and medium vendors to develop the AI solution. The vendors developed a dashboard to deliver the smart CCTV outputs.
The project applied CRISP-DM to cover all vital phases from data explorative analysis, modelling to evaluation and deployment. The solution consists of an AI engine that hosts deep convolution models, covering object detection, object tracking and scene text recognition.
The engine is written in Python, the communication interface is implemented in MQTT and the model's architecture is being selected and configured as per the use case requirement. The models are being custom trained and/or transfer trained as required for the relevant use cases.
Digital and customer capabilities
Ecosystem Partnerships, Growth Mindset, Collaboration and Agility, and Fail Friendly Leadership to develop a proof of concept
In addition to partnering with Cisco for their technical skills, the project team collaborated extensively with various teams within TfNSW and the Commission to understand the business and data needs, address safety concerns and better understand taxi driver behaviours. Consultation with field teams led to insights on which locations could be targeted to deliver maximum benefit for customers.
Fail friendly leadership supported the trial as this idea had not been tested previously. Leaders encouraged trialing different potential solutions based on technological capabilities and encouraged the team to make decisions based on learnings made throughout the process. Agile thinking was encouraged as project scope changed based upon those learnings.
Data Informed Leadership and risk management, to translate and integrate the data captured into a dashboard that met the needs of the business
The project team worked with Cisco and partnered with additional small and medium vendors to design a data interface that translates the images captured into a dashboard. The dashboard displays the data using visual representations, making the data easy to interpret. From a risk management perspective, this ensured that the AI blurred out the faces of customers to protect their identity.
Talent Development, Data Literacy, Ethical Leadership to continue to teach the AI.
During the collaboration with industry, the project team developed their data literacy skills and applied the machine learning skills they learnt during their partnership with Cisco. The project team continued to “teach” the AI and gained an understanding of the types of behaviours the AI can learn, to ensure the images are being interpreted by the AI algorithms appropriately.
Community Engagement, Information access and privacy of citizen data, ethics and social license to engage the community, and protect the privacy of the citizen
The project team engaged with businesses, customers, and local government throughout the project. A combination of direct and electronic communications streams was used to keep stakeholders informed about the project and communicate the strict privacy measures that were applied. Ongoing privacy assessments were conducted for the Privacy Commissioner during both the proof of concept and trial phases and new privacy guidelines for Transport for NSW have been developed for the use of AI including how data captured is collected and shared.
Conclusion
AI can support decision making to provide better outcomes for customers, however, AI cannot do this alone. AI needs humans to apply capabilities and skills such as virtual collaboration, human centred design and machine learning to do this.
The NSW Government is committed to upskilling its workforce and will continue collaborating with the best in the industry so it can respond proactively to the changing nature of work, so we can provide better outcomes for customers.
Complementing the NSW Public Sector Capability Framework which contains the core capabilities for the NSW public sector, the NSW Public Service Commission has developed the Digital and Customer Capability Framework which identifies the six digital and customers capabilities requiring immediate uplift across the NSW public sector to support the Premier’s Priority of being a world-class public service.
The NSW Public Service Commission and Department of Customer Service collaborated with Transport for NSW and the Point to Point Transport Commission on this case study to identify the capabilities and skills required to work with AI and to showcase how we are building a digital and customer capable public sector workforce.