MULTI-SENSOR TRAFFIC DATA FUSION FOR CONGESTION DETECTION AND TRACKING

Abstract

Traffic management applications including congestion detection and tracking rely on traffic from multiple sources to model the traffic conditions. The sources are either stationary sensors which include inductive loop detectors (ILD), radar stations and Bluetooth/WiFi/BLE sensors or Floating Car Data (FCD) from moving vehicles which transmit their locations and speeds. The different sources have their inherent strengths and weaknesses but when used together, they have the potential to provide traffic information with increased robustness. Multi-sensor data fusion has the potential to enhance the estimation of traffic state in real-time by reducing the uncertainty of individual sources, extending the temporal and spatial coverage and increasing the confidence of data inputs. In this study, we fuse data from different FCD providers to improve travel time and average segment speeds estimation. We use data from INRIX, HERE and TomTom FCD commercial services and fuse the speeds based on their confidence values and granularity on virtual sub-segments of 250 m. Speeds differences between each pair of datasets are evaluated by calculating the absolute mean and standard deviation of differences. The evaluation of systematic differences is also performed for peak periods depending on the day of the week. INRIX FCD speeds are compared with ground truth spot speeds where both datasets are measured at a 1-minute interval which show good agreement with an error rate of between 8–20%. Some issues that affect FCD accuracy which include data availability and reliability problems are identified and discussed.