Design and Development of a Novel Framework for the Identification of Roundabouts in a given Map

Abstract

We improve autonomous vehicle navigation by enhancing roundabout detection accuracy. Our streamlined method integrates fundamental map data with existing techniques, simplifying machine learning and achieving precise results. We focus on roundabout detection, junction identification, curved road approximation, road kink detection, and roundabout size validation. The first scheme centers on extracting vital geometric data from map point descriptions to identify map features, including roundabouts. The approach has two components: an algorithm successfully detects roundabouts in 80% of cases, and a machine learning model achieves exceptional accuracy for the remaining cases. Overall, the combined system yields a roundabout detection rate of over 97%. A novel scheme identifies roundabouts in the Map domain by extracting closed loops in roads and using a deterministic algorithm to find common junctions. Data from America and Europe are used to extract features using logical and domain knowledge. The machine model achieves 81% accuracy, reducing manual efforts in verification and offering practical solutions for road mapping. Two adaptive algorithms are proposed to convert image curves into continuous straight lines or polylines, addressing limitations of existing approaches in map-related scenarios. The algorithms use parameterized equations with a tuning value (τ ) to achieve customizable results, showcasing improved performance compared to other methods. The solution offers tunability and enhanced curve approximation. The fourth scheme addresses a practical open problem: identifying kinks in 2D road maps formed by linear collections of lines. Kinks are inversely related to the number of lines, so more lines mean fewer kinks. Our proposed technique utilizes spline interpolation with length manipulation to pinpoint these kink locations. This method is unique, as no other practical technique currently exists for identifying kinks in real map data. With a 10cm threshold, the technique achieved 91% identification accuracy for Frankfurt and 83% for the USA. The proposed Fifth Scheme validates roundabouts through geometry and a spline technique, using a non-rational uniform B-spline function. It calculates the radius based on curvature, fits an ellipse or circle to the shape within a bounding box, and achieves a 92.2% success rate in Europe with a 10-meter threshold. Time saved on manual verification and correction.