Providing more regular road signs infrastructure updates for connected driving: A crowdsourced approach with clustering and confidence level

• Published in: Decision Support Systems Vol. 141; p. 113443
• Main Authors: Tchuente, Dieudonné, Senninger, Dominik, Pietsch, Holger, Gasdzik, Danilo
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-02-2021; Elsevier Sequoia S.A
• Subjects: Automobile industry; Bayesian probabilities; Big data; Cameras; Clustering; Confidence intervals; Connected driving; Crowdsourcing; Digital mapping; Digital maps; Driving; In vehicle; Infrastructure; Intelligent transportation systems; Obstructions; Pavement markings; Road maintenance; Road signs; Satellite navigation systems; Sensors; Street signs; Traffic signals; Traffic signs; Weather; Crowdsourcing; Road signs; Big data; Clustering; Intelligent transportation systems; Connected driving; Bayesian probabilities
• DOI: 10.1016/j.dss.2020.113443
• EISSN: 1873-5797
• ISSN: 0167-9236

Road signs, such as traffic signs, traffic lights or pavement markings, are essential elements for the regulation of driving. Sensors embedded in vehicles (e.g., cameras) are increasingly able to detect them to provide near real-time assistance to the driver, with features such as the current speed limitation at any moment. When sensors are not able to detect road signs (e.g., because of bad weather conditions or obstructions on the road), these features usually rely on in-vehicle digital map layers. However, in-vehicle digital maps are not often up-to-date because their update cycles from map providers are often lengthy (at the scale of several months). For example, a new speed limitation on a road can sometimes take months to be reflected in the vehicle's digital map. To solve this problem, a crowdsourced process that can be used to provide more regular in-vehicle digital map updates (on an hourly or daily basis, for example) is proposed. In this paper, we focus on a crucial step in this process that consists of tracking road sign infrastructure changes by incrementally consolidating crowdsourced cameras' detections of road signs and computing the real positions of the signs, while removing noise due to the imprecision of GPS positions in addition to false positive and false negative detections. This goal is achieved by using non-supervised geospatial clustering techniques and Bayesian probabilities to compute existence probabilities for road signs over time. Overall, this computation is performed in a big data context while also addressing security, privacy, safety and scalability issues. As a proof of concept, two experiments are conducted with true field data and they clearly demonstrate the relevance of the proposed approach. The method or the platform can be useful for many market players such as car manufacturers, map providers, or GPS providers (including navigation software providers) to provide more frequent map updates, to make connected driving easier and safer. It can also be useful for road infrastructure maintenance by helping to identify road signs that are poorly positioned or are not very visible.
•Providing up-to-date and accurate road signs positions from crowdsourced camera's detections and car's positions.•Possible impacts on reducing in-car digital map updates frequencies or identifying misplaced road signs.•Algorithm with adapted clustering techniques and confidence level based on Bayesian probabilities and exponential decay.•Relevance of the approach with two experimentations in real driving conditions and with true field data.