Design, Implementation, and Evaluation of Radar Sensor Models in a Virtual Environement

Abstract

Abstract This project aims to design, implement, and evaluate radar sensor models to be used in a virtual environment as the current virtual sensor model lacks realistic radar properties such as measurement error and detection rate. Data driven models were designed using collected data measurements from testing of the forward looking radar. The testing was conducted with a robot mounted with a pedestrian child doll driving in a fan and circular pattern. New field of view models were designed as a concave and convex polygon as well as a polynomial consisting of three polynomial equations. Analyzing the fan pattern measurement error showed that small areas had a similar magnitude and direction. Therefore, a segment grid similar to the testing grid was constructed in which the mean and variance were calculated. The model samples an error from the normal distribution in each segment. Furthermore, the measurement error model calculates an error depending on the target trajectory. The detection rate model calculates the probability of detection from the number of detections and misses in each segment. Each model was evaluated based on their accuracy, execution time, and repeatability. The evaluation showed that the concave model was the most optimal when evaluating the execution time and field of view. The measurement error model did not affect the number of detections and correctly calculated the error in the fan and circle pattern, further data collection is needed to verify other driving trajectories. The detection rate model altered the field of view based on the probability of detections in each segment.