Statistical modelling of critical cut-in trajectories based on naturalistic driving data

Examensarbete för masterexamen

Please use this identifier to cite or link to this item: `https://hdl.handle.net/20.500.12380/304252`
File Description SizeFormat
2021-37 Sainath Kishore Josyula.pdfMaster thesis1.98 MBAdobe PDF
Bibliographical item details
FieldValue
Type: Examensarbete för masterexamen
Title: Statistical modelling of critical cut-in trajectories based on naturalistic driving data
Authors:
Abstract: Cut-in maneuvers are events when a vehicle changes lane and moves close to another vehicle in the adjacent lane. This phenomenon is quite common on highways and has adverse impact on traffic safety. Statistics from the Fatality Analysis Reporting System (FARS) by National Highway Traffic Safety Administration (NHTSA) show that there have been more than 290,000 traffic crash injuries associated with cut-in maneuvers (including rear-end, angle or sideswipe collision) between years 2015 and 2019. Active safety systems and autonomous vehicles are being developed to achieve safe driving and should be able to detect potentially dangerous scenarios like critical cut-ins, and act to avoid them. Statistical models of cut-in scenario trajectories are useful for developing and evaluating both active safety systems and autonomous vehicles. This thesis aims to increase our understand of and model cut-in trajectories of vehicles performing cut-in maneuvers, using SHRP2 (The Second Strategic Highway Research Program) naturalistic driving data. To conduct the study, the SHRP2 event data has been manually categorized. Thereafter, a dataset of kinematic variables, which were extracted using a video annotation tool, has been prepared to enable studying the trajectories in detail. Specifically, this study uses a quintic polynomial of time to model lateral and longitudinal trajectories of the vehicle that cuts in (or principle another vehicle; POV). One of the required inputs is the event duration which is calculated by identifying maneuver start and end times . The event durations follow a normal distribution and range from 2.1 to 6.4 seconds. Then, two linear models of event duration are built to calculate the remaining two variables required for a polynomial trajectory model, which are initial POV lateral acceleration and final POV longitudinal position. Using these three variables a range of trajectories have been generated, representing SHRP2 naturalistic driving for right to left single lane change cut-ins. This thesis also uses a probabilistic regression model to calculate the distribution of parameters of quintic polynomial of lateral position of POV. Using this model, new (other than training data) trajectories have been generated. Both these generative models of trajectories can be used as one of the inputs to simulations used in the design and evaluation of active safety systems and automated driving systems.
Keywords: Autonomous Driving;Cut-in maneuvers;Trajectory models;Naturalistic driving data;Probabilistic regression
Issue Date: 2021
Publisher: Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper
Series/Report no.: 2021:37
URI: https://hdl.handle.net/20.500.12380/304252
Collection:Examensarbeten för masterexamen // Master Theses