Sensor Device on Lateral Passing Distance: A Case Study With Unsupervised Learning Model to Estimate Hazardous Areas for Active Mobility
Researcher Luiz Marcel Silva de Mello
ABSTRACT
Commuting by bicycle is widely increasing worldwide. As an active transportation mode, cycling can potentially reduce traffic congestion and air pollution. Also, promoting an active lifestyle can improve public health and make cities more human-friendly. Although, the quantity of occurrences and fatalities with cyclists is still worrisome.
The Surrogate Safety Measures (SSM) are promising indicators for assessing traffic safety with measures based on these traffic conflicts. The word “surrogate” is used because the measures are based not on crashes but on traffic conflicts. Also, network screening ensures an efficient identification of hazardous sites to reduce the number and severity of crashes. This methodology can be conducted using either a reactive or a proactive approach. Regarding the proactive approaches, bicycles instrumented with sensors became increasingly usable for research in the mobility field. By using a portable and multi-functional sensing device is possible to collect bicycle trajectory data and Lateral Passing Distance (LPD) using various sensors connected to a database system.
Therefore, the current research aims to estimate and define hazardous areas for active mobility by applying unsupervised machine learning algorithms (k-means and DBSCAN) based on a sensor device for data collection. The Lateral Passing Distance (LPD) results collected between bicycles and vehicles were related to the cyclist data. Beyond the clustering investigation, a correlation between the features has identified how the data interacted among them.
Some of this data includes velocity, curse elevation, altitude, accelerometer, and gyroscopic information from a naturalistic data collection on the street. Regarding the general data, 25% of the readings are less than 139.62cm. When the clustering model was applied, 25% of the LPD readings were less than 100.13cm; for the second quartile, 50% were less than 193.69cm. It indicates critical LPD for one of the clusters with 75 readings, considering the threshold of 150cm for the minimal lateral clearance distance law adopted in Brazil. The methodology was applied to a case study regarding the Brasilia city center avenue with a shared pathway around the local City Park. Therefore, this study aims to propose a methodological and data-driven approach to bicycle safety using machine learning algorithms.
Keywords: Bicycle Safety; Lateral Passing Distance; Machine Learning; Clustering Model; Sensor; Intelligent Transportation System; Road Safety; K-Means; DBSCAN.