Surrounding Vehicle Trajectory Prediction Using Lane Selection Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in accurately predicting the future trajectory of surrounding vehicles in autonomous driving, particularly due to difficulties in fusing image-based environment information with coordinate-based past trajectory information.
Innovation Solution
The proposed solution involves an apparatus and method that utilize a Lane Selection Network (LSN) and a Trajectory Prediction Network (TPN) to predict the future trajectory of a surrounding vehicle. This involves inputting past trajectory information and lane information into the LSN to detect reference lane information, which is then inputted into the TPN along with the past trajectory information to output future trajectory information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image-based environment information and coordinate-based past trajectory information are fused using existing technologies, then trajectory prediction can be performed, but the fusion is difficult and accuracy is degraded
Solution Approach 1:
The patent transforms image-based environment information into coordinate-based road information by detecting lane lines and converting them into lane coordinate vectors. This parameter transformation allows both image information and trajectory information to be processed in the same coordinate system, enabling accurate fusion without complex multi-modal processing.
Solution Approach 2:
The patent introduces road information (lane coordinate vectors) as an intermediary that bridges image-based environment information and coordinate-based trajectory information. This intermediary representation in a unified coordinate system facilitates seamless fusion of the two data types.
2Measurement precision
If only object information from surrounding environment images is reflected, then processing is simplified, but trajectory prediction accuracy is degraded
Solution Approach 1:
The patent extracts road information (lane lines, road boundaries) from the surrounding environment images, separating this critical structural information from general object detection. This extracted road information is then used as a foundation for accurate trajectory prediction, going beyond mere object detection.
Solution Approach 2:
The patent segments the environment information processing into distinct components: road information extraction (lane line detection), object information detection, and trajectory analysis. This segmentation allows each component to be processed optimally and fused effectively.
3Reliability
If high definition map and past trajectory information are used for prediction, then some accuracy is achieved, but the prediction accuracy remains insufficient for safe autonomous driving
Solution Approach 1:
The patent merges multiple information sources including high definition map data, detected road information from images, past trajectory information, and object information into a unified prediction framework. This comprehensive integration of diverse data sources achieves the high accuracy required for safe autonomous driving.
Data Source
AI summary
An apparatus for predicting a trajectory of a surrounding vehicle includes a storage configured to store a high definition map, a lane Selection Network (LSN), and a Trajectory Prediction Network (TPN) and a controller that extracts lane information around a target vehicle, traveling around a host vehicle, based on the high definition map, inputs the lane information around the target vehicle and previous trajectory information of the target vehicle to the LSN to detect reference lane information, and inputs the reference lane information and the previous trajectory information of the target vehicle to the TPN to acquire future trajectory information of the target vehicle.