High Precision Map Conflation of Fleet Sourced Traffic Signs

The ever-increasing demand for digital maps in various domains amplifies the importance of having accurate and up-to-date maps. To address this, the proposed system pervasively conflates large volume of sign detections recorded by a transportation fleet of vehicles into map database. Detected and geo-localized sign objects collected from the fleet over a time period are passed through a context-aware clustering for aggregation and followed by map matching with a new Hidden Markov Model (HMM) that utilizes vehicle GPS and compass sensors. Eventually, a resolution model utilizes features from detection, traversal, and surrounding map context to assign a confidence score for identified new signs for ingestion via a rapid resolution route. On a test data across the USA geography with large volume of detections, the proposed system could add 51% stop signs and 21% of traffic lights to the map at average precision of 99.55% via rapid resolution path.