Automatic detection of traffic lights, street crossings and urban roundabouts combining outlier detection and deep learning classification techniques based on GPS traces while driving

Mario Muñoz Organero, Ramona Ruiz-Blazquez, Luis Sánchez Fernández

Abstract: Highlights•We provide a complementary method to current road element detection algorithms with minimal requirements (only GPS data).•Both classification and detection algorithms are shown using a novel combination of outlier detection and machine learning.•A novel intra and inter-drive outlier detection schema divides road infrastructural elements from sporadic traffic incidents.•Using deep learning, speed and acceleration patterns are analyzed at each outlier to extract relevant features.•By adding the degree of atypicity for each point, the algorithm achieves a recall of 0.89 and a precision of 0.88.