CycleAR: Exploring Augmented Reality for Safer and More Immersive Cycling Experiences

Cycling in urban environments can often be intimidating, particularly in areas with dense traffic and limited cycling infrastructure. The CycleAR project, currently under development within the Department of Systems and Control Engineering at the University of Malta, aims to explore how augmented reality (AR) and immersive simulation technologies may contribute towards safer and more accessible cycling experiences.

CycleAR is investigating the development of an AR-based cycling simulation platform that reconstructs realistic cycling environments from real-world video footage. By combining techniques from computer vision, image processing, and immersive visualisation, the project seeks to create a system that allows cyclists to experience realistic road scenarios within a controlled virtual environment.

Unlike many existing cycling simulators that rely on artificial or game-like environments, CycleAR focuses on the cyclist’s real perspective and on real road conditions. The long-term vision is to allow cyclists to train virtually on routes that closely resemble real urban environments, with particular emphasis on Maltese roads and traffic scenarios.

The project makes use of video footage captured from cyclist-mounted cameras and investigates how fragmented recordings may be stitched together into continuous navigable routes. The work also explores the use of style-transfer techniques to harmonise lighting conditions between different video segments and monocular depth estimation methods to reconstruct three-dimensional scenes from standard video footage.

Since the video data is being collected from live urban streets, particular care is also being taken to address ethical and privacy considerations associated with the recording of public environments. The project is therefore incorporating automated privacy-preserving mechanisms that identify and blur vehicle number plates prior to further processing of the video data. This is being achieved through a combination of vehicle detection and number-plate localisation techniques implemented using YOLO-based object detection models as shown in Figure 1.

While the automated system performs well in many situations, some challenging cases still require manual intervention. These include vehicles captured at highly oblique viewing angles — such as parked cars appearing in the peripheral field of view — or situations where only partial number plates are visible within the scene. In such cases, manual annotation is being used to ensure that identifiable information is consistently obscured.

Figure 1: Number plate anonymisation. The figure on the right shows the result of the car detection (green) and the detected number plates (red). While the figure on the left shows the result with the number plates blurred. Note that for purpose of this diagram, the number plates in the left image have been manually blackened out.

Similarly, the project also plans to implement face detection and face-blurring techniques to preserve the privacy of pedestrians and other individuals appearing in the captured footage. These measures form an important component of the research pipeline, ensuring that the development of immersive AR environments remains aligned with ethical data-handling practices and privacy protection principles.

While the project is still in its early stages, several proof-of-concept components are already being assembled. One of the first practical developments involved converting a regular bicycle into a static training bicycle through the integration of a SARIS wheel trainer. The system currently captures wheel-speed information from the trainer via Bluetooth communication, allowing the rider’s pedalling speed to control the movement through the virtual scene. To validate the measurements being obtained, the data from the trainer has also been compared against readings from an independent speed sensor mounted directly on the bicycle wheel.

Early experimentation has highlighted one of the practical challenges associated with real-time immersive control systems. Although the Bluetooth communication provides continuous speed updates, the transmitted signal can contain fluctuations and short-term variations that make scene navigation appear less natural, causing the virtual environment to advance in intermittent fits and starts rather than through smooth continuous motion. To address this, the project is currently investigating data smoothing and filtering techniques to stabilise the control signal and produce more fluid and realistic progression through the virtual environment as shown in Figure 2.

Figure 2: A Magene speed sensor mounted on the bicycle wheel used to compare and verify the speed data obtained from the Saris wheel trainer. The speed data can from the sensor is smoothened to obtain a more stable signal to obtain better control of the virtual environment.

This creates a more natural interaction model, where the bicycle itself determines the speed of progression through the simulated environment rather than relying on a conventional joystick-based interface.

The project is now entering an exciting next phase, where reconstructed video scenes and virtual environments will begin to be projected seamlessly within the Department’s immersive CAVE (Cave Automatic Virtual Environment) system. The department’s CAVE facility provides a large-scale immersive projection environment capable of surrounding the user with the virtual scene, opening new possibilities for realistic cyclist-perspective simulations and interactive experimentation.

Beyond the immediate technical goals, CycleAR also highlights the broader potential of AR technologies in areas such as road safety research, urban mobility, behaviour analysis, and active transport promotion. By allowing users to experience complex traffic situations within a safe and controlled environment, such systems may eventually contribute towards cyclist training, awareness initiatives, and even urban planning studies.

The project remains a work in progress, with ongoing research focusing on scene reconstruction, immersive rendering, sensor integration, user interaction, and privacy-preserving video processing. Nevertheless, the early developments already demonstrate the exciting possibilities that emerge when augmented reality, computer vision, and human-centred simulation technologies are brought together within a real-world application context.

---

The CycleAR project is funded by Xjenza Malta through the Research Excellence Program with research grant REP-2025-027.

The project investigators are Prof. Alexandra Bonnici, Prof. Kenneth Camilleri, Dr Stefania Cristina and Mr Samwel Portelli.