In English

Tool development and quantitative analysis for naturalistic Left Turn Across Path/Opposite direction (LTAP/OD) driving scenarios

Lin Meng ; Jifeng Wang
Göteborg : Chalmers tekniska högskola, 2016. Diploma work - Department of Applied Mechanics, Chalmers University of Technology, Göteborg, Sweden, ISSN 1652-8557; 2016:53, 2016.
[Examensarbete på avancerad nivå]

This thesis aims to develop a tool to obtain key variables of other vehicles from video and apply this tool for the quantitative analysis of driver behaviour for Left Turn Across Path/Opposite Direction scenarios. The variables include relative speed and relative positions between subject vehicle and oncoming vehicle. Three methods are discussed and implemented in software tool for manual annotation,: a ground points method, a vehicle width method and an optical flow matching method In addition Kalman filter is applied to integrate this three methods together with a constant acceleration model. An experiment shows the range estimation result has an average percentage error of less than 10%, within the range 10m to 50m, and that the speed estimation has around a 10% error at approximately 10m and 20% error around 20m. Semi-automatic methods for extracting the desired variables is also presented. Based on manually selected tracking region in the first frame, and optical flow computed through the video, the desired (manually selected) region can be tracked. Optical flow vectors on the region has a relationship with motion. Motion estimation is accomplished with a matching process. After applying the tool on 102 LTAP/OD cases in a subset of EuroFOT data, the Post Encroachment Time was calculated for each. Results show that drivers feel comfortable to turn into the encroachment zone in a range between 2 and 4s after the last oncoming vehicle leaves that zone.

Nyckelord: optical flow, vehicle kinematics, naturalistic driving data, LTAP/OD, manual annotation



Publikationen registrerades 2016-10-25. Den ändrades senast 2016-10-25

CPL ID: 244211

Detta är en tjänst från Chalmers bibliotek