Vehicle Trajectory Prediction Using Rasterized Multimodal Environment Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing trajectory prediction methods for autonomous driving fail to ensure safety in complex and diverse driving scenarios due to the inability to fully account for environmental multimodality and uncertainty.
Innovation Solution
A method and apparatus that utilize a neural network prediction model to process raster image data, generating a multidimensional feature map that considers ego-vehicle and environmental data, ensuring the safety of autonomous driving by predicting vehicle trajectories in complex scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional trajectory prediction methods are used, then the system is simple and easy to implement, but the prediction accuracy and safety are insufficient in complex driving scenarios
Solution Approach 1:
The prediction system is segmented into multiple specialized modules: raster image generation module, neural network processing module, multidimensional feature map generation module, and trajectory prediction module. Each module handles specific aspects of the prediction task, improving overall reliability while managing complexity through functional decomposition.
Solution Approach 2:
The system transforms traditional 2D sensor data into 3D raster image data with spatial-temporal dimensions. It generates multidimensional feature maps that incorporate multiple modalities (occupancy grid, heatmap, trajectory vectors) simultaneously, enabling more comprehensive environmental understanding and safer predictions.
2Measurement precision
If comprehensive environmental data is processed to account for multimodality and uncertainty, then the prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The system performs preliminary actions by pre-processing sensor data into structured raster images with embedded spatial-temporal relationships before neural network processing. It pre-generates occupancy grid maps and heatmap representations that organize environmental data in advance, reducing the computational burden during real-time prediction while maintaining high measurement precision.
Solution Approach 2:
The neural network acts as an intermediary that processes complex multidimensional feature maps and transforms them into actionable trajectory predictions. This intermediary layer manages the complexity of processing multiple environmental modalities while preserving measurement precision through learned feature representations.
3Loss of information
If raster image data with spatial-temporal environment is used, then the comprehensive environmental understanding improves, but the data processing time increases
Solution Approach 1:
The system performs preliminary actions by pre-organizing environmental sensor data into compact raster image formats that encode spatial-temporal relationships efficiently. This pre-processing structure enables the neural network to quickly access and process comprehensive environmental information without excessive processing time during real-time prediction.
Solution Approach 2:
The system changes the parameter representation of environmental data from traditional coordinate-based formats to raster image parameters with fixed dimensions and structured channels. This parameter transformation enables more efficient processing while preserving complete environmental information through the raster encoding scheme.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Disclosed are a method and an apparatus for predicting vehicle trajectory, and a method and an apparatus for training a neural network prediction model, relating to the field of intelligent driving technology. The method for predicting vehicle trajectory including: determining raster image data of a vehicle at a current time point based on ego-vehicle travelling data of the vehicle during travelling and environmental data of the vehicle; processing the raster image data based on a neural network prediction model, and predicting a trajectory of the vehicle in a future predetermined time period based on the multidimensional feature map, ego-vehicle size data and the environmental data.