Author ORCID Identifier

https://orcid.org/0009-0002-5612-7775

Semester

Fall

Date of Graduation

2023

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Brian Woerner

Committee Member

Andrew Nix

Committee Member

Natalia Schmid

Abstract

This work explores the implementation of sensor fusion and data association for autonomous vehicle design. Advancements in Adaptive Driver Assistance System (ADAS) technology have driven the development of perception algorithms required for higher levels of autonomy in vehicles. Perception algorithms process data collected from radar, camera, and LiDAR sensors to generate a complete model of the ego vehicle’s surrounding environment. Fusion of data from these sensors is important for accurate measurement of longitudinal and lateral distances to surrounding objects. Sensor fusion associates sensor detections to each other through different data association techniques. Data association techniques can consist of independent assignment of sensor detections, the Hungarian algorithm, or clustering algorithms such as Fuzzy C-Means (FCM). One baseline sensor fusion technique is a simple weighted average, which can yield satisfactory accuracy. The goal of this work is to evaluate the performance of sensor fusion utilizing a weighted average with an advanced Fuzzy C-Means data association algorithm. The results are applied to a modified Chevrolet Blazer, used by WVU for the EcoCAR Mobility Challenge (EMC) Year 4 competition. The secondary goal of this work is to implement FCM for the use of data association and compare the performance to the same initial sensor fusion design. For weighted average testing, real-world sensor data from the Intel Mobileye 630 camera and Bosch Mid-Range Radar are used to evaluate different static weights. The results from the static weights are utilized to create dynamic weights for the weighted average and the performance of static weights and dynamic weights are compared. For data association testing, simulated sensor data from camera and radar detection models are used to evaluate the detection association performance of the baseline sensor fusion to FCM implemented sensor fusion. Results show dynamic weights improved the baseline sensor fusion’s longitudinal distance error by 6.80% for approach tests and 5.21% for departure tests. Results for data association testing showed the baseline sensor fusion had an average accuracy of 66.65% and FCM implemented had 51.65% for 100% probability of detection, but for 25% probability of detection, baseline was 20.19% and FCM was 40.81%. Recommendations are made to improve the performance of the weighted average for more accuracy in longitudinal distance and expanding the FCM research to utilizing real-world sensor data.

Download

Share

COinS