Analysing cyclists’ data is paramount to foster cycling as a mode of transport and to help urban planners and decision makers to better design safer cycling infrastructures. In fact, with a growing trend on both data driven approaches and cycling as a mode of transport in cities, we believe that one approach to better understand cycling safety is by looking at what bicyclers face on a daily setting.
With this in mind, we propose an automatic event recognition system capable of classifying cycling events using accelerometer and gyroscope data from a smartphone. Data was collected under real-world conditions and without any supervision from any researcher. Subjects were instructed to record a set of 7 different cycling events using a smartphone without any restriction, i.e., the smartphone could be placed anywhere (bicycle handle, pants’ pocket, jacket’s pocket, …). Thus, from inertial data (accelerometer and gyroscope) recorded using a smartphone we propose a method for automatically identify and recognize different activities, such as cycling maneuvers. We use Support Vector Machines (SVMs) to classify different events such as Normal Riding, Braking, Stopped, Turning Smoothly or Turning Abruptly. We test our methods on real data, attaining an accuracy of about 85% on maneuver classification, which are promising results.