Date of Graduation


Document Type


Degree Type



Statler College of Engineering and Mineral Resources


Lane Department of Computer Science and Electrical Engineering

Committee Chair

Brian Woerner

Committee Co-Chair

Andrew Nix

Committee Member

Andrew Nix

Committee Member

Thomas Devine

Committee Member

Roy Nutter

Committee Member

Gianfranco Doretto


Sensor fusion is a key system in Advanced Driver Assistance Systems, ADAS. The perfor-
mance of the sensor fusion depends on many factors such as the sensors used, the kinematic
model used in the Extended Kalman Filter, EKF, the motion of the vehicles, the type of
road, the density of vehicles, and the gating methods. The interactions between parameters
and the extent to which individual parameters contribute to the overall accuracy of a sensor
fusion system can be difficult to assess.
In this study, a full-factorial experimental evaluation of a sensor fusion system based
on a real vehicle was performed. The experimental results for different driving scenarios
and parameters are discussed and the factors that make the most impact are identified.
The performance of sensor fusion depends on many factors such as the sensors used, the
kinematic model used in the Extended Kalman Filter (EKF) motion of the vehicles, type of
road, density of vehicles, and gating methods.
This study identified that the distance between the vehicles has the largest impact on the
estimation error because the vision sensor performs poorly with increased distance. In addi-
tion, it was identified that the kinematic models had no significant impact on the estimation.
Last but not least, the ellipsoid gates performed better than rectangular gates.
In addition, we propose a new gating algorithm called an angular gate. This algorithm
is based on the observation that the data for each target lies in the direction of that target.
Therefore, the angle and the range can be used for setting up a two-level gating approach
that is both more intuitive and computationally faster than ellipsoid gates. The angular
gates can achieve a speedup factor of up to 2.27 compared to ellipsoid gates.
Furthermore, we provide time complexity analysis of angular gates, ellipsoid gates, and
rectangular gates demonstrating the theoretical reasons why angular gates perform better.
Last, we evaluated the performance of the Munkres algorithm using a full factorial design
and identified that narrower gates can speedup the running time of the Munkres algorithm
and, surprisingly, even improve the RMSE in some cases.
The low target maneuvering index of vehicular systems was identified as the reason why
the kinematic models do not have an impact on the estimation. This finding supports the use
of simpler and computationally inexpensive filters instead of complex Interacting Multiple
Model filters. The angular gates also improve the computational efficiency of the overall
sensor fusion system making them suitable for vehicular application as well as for embedded
systems and robotics.