Autonomous Vehicle Perception System for Adverse Weather Lane Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous driving vehicles face challenges in reliable lane and object detection, particularly in adverse weather conditions where LIDAR sensors are unreliable, necessitating a vision-based perception system for safe and efficient navigation.
Innovation Solution
A multi-stage perception workflow using single or multiple front-facing cameras and a frontal radar sensor, combined with machine learning models, to preprocess, process, and post-process images for lane and object detection, enabling autonomous driving even without map information and improving performance over time with data collection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LIDAR sensors are used for lane and object detection, then detection accuracy is improved under clear weather conditions, but reliability deteriorates in rain or snow conditions
Solution Approach 1:
The patent changes the detection parameter from LIDAR (light-based) to EO/IR camera sensors (electromagnetic radiation detection) that can operate effectively in various weather conditions including rain and snow, thereby maintaining reliability while adapting to adverse environmental parameters
Solution Approach 2:
The patent uses EO/IR camera sensors to create visual copies/images of the environment that can be processed to detect lanes and objects, providing an alternative sensing mechanism that replicates the detection function without relying on LIDAR's light reflection principle that fails in adverse weather
2Reliability
If vision-based perception systems are used instead of LIDAR, then reliability in adverse weather is improved, but measurement precision may deteriorate
Solution Approach 1:
The patent introduces machine learning models as intermediary processing layers that enhance the EO/IR camera data, extracting and refining lane and object information to achieve measurement precision comparable to or exceeding LIDAR while maintaining weather reliability
Solution Approach 2:
The system performs preliminary processing of EO/IR images through machine learning models before final detection, pre-enhancing the data quality and feature extraction to ensure high measurement precision is achieved in the detection stage despite using vision-based sensing
3Reliability
If multiple sensors and processing stages are implemented, then detection reliability is improved, but device complexity increases
Solution Approach 1:
The patent segments the perception system into distinct functional modules (EO/IR sensing, machine learning processing, lane detection, object detection) that can be developed, tested, and maintained independently, managing complexity through modular architecture while achieving high reliability through integrated operation
Data Source
AI summary
According to some embodiments, a system pre-processes, via a first thread, a captured image perceiving an environment surrounding the ADV obtained from an image capturing device of the ADV. The system processes, via a second thread, the pre-processed image with a corresponding depth image captured by a ranging device of the ADV using a machine learning model to detect vehicle lanes. The system post-processes, via a third thread, the detected vehicle lanes to track the vehicle lanes relative to the ADV. The system generates a trajectory based on a lane line of the tracked vehicle lanes to control the ADV autonomously according to the trajectory.