Research Question/Objective: Many bicyclists were killed or injured when they were riding on the street because of the traffic crashes. To avoid or mitigate accidents, Pre-Collision System (PCS) is introduced to give drivers warning or make the car brake automatically. Analysis of the bicycle speed for assessment of bicyclist Pre-Collision System is critical in developing realistic test scenarios, which can also help design PCS evaluation system and improve PCS itself. In this study, bicycle speeds were analyzed in two scenarios including “along the road” and “across the road”. Furthermore, one special case we called “ride out” in crossing the road was analyzed separately because of the inconstant speed during the whole crossing period. All analysis results provide reference for developing bicyclist Pre-Collision System evaluation and are used by real testing scenarios in Transportation Active Safety Institute(TASI) lab at IUPUI.
Methods and Data Sources: Bicyclist information was taken from TASI 110-car naturalistic driving video database which was collected in the metropolitan area of Indianapolis. We obtained GPS locations and corresponding time stamps of the bicyclists through an image-processing-based semi-automatic process for a subset of the overall database. For traveling along the road cases, bicyclist speed was calculated based on the changes in GPS locations and the corresponding traveling time. For crossing cases, three to five velocity values were achieved during the whole period. These values are proved to be consistent for “ride-through” crossing scenario. To best fit speed situation for “ride-out” scenarios, we built a speed model including two consecutive stages, accelerating stage and constant-speed traveling stage. Then the optimized traveling trajectory was achieved using cost function method in MATLAB after a two-way search when dividing the ride- out crossing process in eight different ways.
Results: Per three main scenarios of the bicycle motion, 895 bicyclist cases were obtained randomly from the database. Statistics analysis results suggest 25 percentiles, mean value, and 75 percentiles of bicyclist traveling speeds in different scenarios. Especially for the ride-out scenario, the speed model we build quantitatively explain and predict the bicyclists speed.
Discussion and Limitations: Comparing to other constant speed or mean speed scenarios, our model is more comprehensive considering the speed variation during the whole motion period. We could more accurately estimate bicyclists speed in different scenarios. However, due to the limit amount of cases, we cannot break down more detail scenarios such as the street conditions or weather conditions. In future research, we may enlarge our sample database to improve our model close to realistic situation more deeply.
Conclusion: Studying information from TASI-110 car naturalistic database, we calculated bicyclist speeds in different situations. We specially focused on the “ride out” situation and analyzed the curve trend of its whole process. These data results were used to develop for bicyclist Pre- Collision System evaluation scenarios. These speed data give a great contribution to Pre-Collision System development and may help other researchers for further research.