Autonomous Vehicle Sensor Perceptual Quality Assessment

Perception is one the most fundamental but most important step for autonomous vehicle (AV). Few studies have been conducted for AV surrounding driving environment complexity and sensor perceptual quality. In this project, we studies the evalution method about driving environment complexity and sensor perceptual quality. Then, we developed deep learning-based algorithms to predict both the driving environment complexity and sensor perceptual quality factors.

Autonomous Vehicle Driving Environment Complexity Perception

paper: Attention-based Neural Network for Driving Environment Complexity Perception

Environment perception is crucial for autonomous vehicle (AV) safety. Most existing AV perception algorithms havenot studiedthe surrounding environment complexity and failed to include the environment complexity parameter. In this project, we propose a novel attention-based neural network model to predict thecomplexity level of thesurrounding driving environment.

The image is processed by a heat map generation algorithm

The surrounding environment complexity label is generated through subjective labeling.

An attention network is devloped to predict the surrounding environment complexity label

The network classification accuracy achieved over 90%

Our new research about sensor perceptual quality is about to be published soon! The future algorithms are going to be open-source in the future. Welcome to visit after the publication!