Medium-Fidelity Evaluation and Modeling for Perception Systems of Intelligent and Connected Vehicles

Abstract

This paper proposes a framework for evaluating and modeling perception systems, motivated by the need to develop testing scenarios for verification and validation of autonomous driving systems operating in various driving environment perception approaches, including both ego-vehicle centric perception and cooperative perception with enabled connectivity. The proposed perception system evaluation and modeling approach is probabilistic, with perception failures and errors encoded as stochastic processes and accounts for the operation domain. The perception error model is parameterized to consider both spatial and temporal aspects in the offline evaluation process. The proposed method exhibits well-fitting performance on the model of the perception error pattern based on evaluation results in various virtual and real traffic data with several benchmark perception algorithms.

Reza Mehrzi

Data Scientist

My research interests include building and developing predictive models using machine learning methods, deep learning object detection and semantic image segmentation, experimental designs, and survival analysis.