99279Pelation REBO: Developing a machine learning approach to automatically identify cycling near misses and root causes from dashcam footageClosedSmall Business Research InitiativeInnovate UKPublic description Pelation is a cycle technology company that aims to eliminate sustainable mobility barriers through innovative design and engineering. Focused on the elimination of dangerous near misses and close passes, Pelation's current product REBO, is an internet connected bike light, dashcam and GPS-enabled bookmark button that allows cyclists to capture previously unavailable near miss footage and information with a click of a button. Current cycling road data on the market give an overview of problem areas but require leap-of-faith inference and assumption on specific incidents. It is difficult to gain confidence in incident root causes without the incident context - our video footage captures this missing information which enables the determination of how near misses develop. The next stage of the technology roadmap is to develop capabilities to automatically identify and analyse cycling near misses using our device's video footage and data. This aims to produce, for these near misses, actionable insights to determine their root causes and potential fixes. This will allow local authorities to easily understand, prioritise, and implement action plans faster and with more impact and value for money. To achieve this, Phase One of Pelation's project is a feasibility study that sets out to research key near miss contributing factors, develop the machine learning approach, and identify additional sensor specifications required to automate the near miss determination and root cause identification process from our devices' footage and sensors. Pelation will be working closely with Oxfordshire County Council on this project - identifying their needs and requirements and verifying current challenge areas. Our novel approach using machine learning to automatically identify near misses from cyclists submitted footage will enable drastically reduced time and cost spent analysing these existing data sets and provide an objective overview of near miss root causes that provide immediately usable and actionable information. This project develops an innovative machine learning model to identify categories and key factors for near misses from cycle footage, and develops a pattern matching algorithm that matches key incident factors to geospatial/kinematic sensors data. Phase Two of the project will be to develop a scalable near miss identification and root cause determination software that will be built into our devices and cloud analysis platform. This will be followed by a large scale road trial (with a variety of people, places, and time) in collaboration with a local authority to demonstrate the usefulness of the technology in real life operating conditions.