A Method for Designing the Perception Module of Autonomous Vehicles Using Stereo Depth and Semantic Segmentation
Fengchen Wei, Weiji Wang · 2024
In this study, a perception module was developed for autonomous driving systems utilizing stereo matching and semantic segmentation models. This module serves two crucial functions for autonomous vehicles. Firstly, it combines object detection and stereo matching techniques to accurately determine distances for both the ego vehicle and target vehicles, thereby informing the control module's strategy selection. Secondly, it employs semantic segmentation models and filtering algorithms to devise a method for calculating drivable areas for autonomous vehicles, which is essential for the subsequent planning module. The modular design of the system allows for flexible training and selection of the backbone network for each module, enhancing adaptability and reducing computational costs during simultaneous training.