Motion Trajectory Forecasting With 3D Pointwise Temporal Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional motion forecasting methods in autonomous driving fail to fully utilize implicit information about object motion trends, resulting in low accuracy of trajectory prediction.
Innovation Solution
The method employs Temporal Point Cloud Networks (TPCN) with dual-representation spatial learning and dynamic temporal learning to extract spatial and temporal pointwise features from past trajectories, integrating time information for improved prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods use kinematic constraints and high-precision map information with rasterizing, then the method can process map and object data, but the accuracy of motion trajectory prediction is low due to not fully utilizing implicit motion trend information
Solution Approach 1:
The patent transforms trajectory data from traditional 2D spatial representation to 3D point cloud representation by adding vertical dimension with time stamps. This dimensional expansion allows the model to capture implicit motion trends more effectively by representing trajectory points as three-dimensional coordinates (x, y, t), enabling better utilization of temporal and spatial information simultaneously.
Solution Approach 2:
The patent changes the parameter representation from rasterized grid cells to continuous point cloud coordinates with temporal dimensions. By transforming the data structure from discrete spatial bins to continuous spatiotemporal points, the model can preserve more implicit motion information and achieve higher prediction accuracy through learned temporal patterns.
2Reliability
If traditional methods apply standard image backbone networks to rasterized data, then the processing pipeline is simple, but the model cannot capture temporal dynamics and motion trends effectively
Solution Approach 1:
The patent replaces traditional image processing mechanics with point cloud processing mechanics. Instead of using convolutional neural networks designed for 2D images, the model employs point cloud-based neural networks that can naturally handle three-dimensional spatiotemporal data, enabling effective capture of temporal dynamics while maintaining architectural efficiency.
Solution Approach 2:
The patent introduces dynamic temporal learning mechanisms that adaptively process varying-length trajectory sequences. The model uses dynamic graph convolutions and temporal attention mechanisms that adjust to different motion patterns and time intervals, enabling reliable temporal dynamics capture while managing complexity through selective activation of computational components.
3Measurement precision
If the model processes detailed trajectory points with temporal features, then prediction accuracy improves, but computational complexity and processing time increase
Solution Approach 1:
The patent segments the trajectory processing into spatial feature extraction and temporal feature extraction stages. By dividing the computational task into separate spatial and temporal processing streams, the model can efficiently handle detailed trajectory points without excessive computational overhead, maintaining high accuracy while reducing processing time through specialized sub-networks.
Solution Approach 2:
The patent applies partial processing by selectively focusing computational resources on critical trajectory points and time intervals. The model uses attention mechanisms to identify and process only the most informative portions of trajectory data, achieving high prediction accuracy without processing every detail equally, thus reducing overall processing time.
Data Source
AI summary
The present disclosure relates to a method for forecasting a motion trajectory, a computer-readable storage medium, and a computer device. The method includes: obtaining an observed past trajectory of an object; obtaining a spatial pointwise feature of each trajectory point in the observed past trajectory; obtaining a temporal pointwise feature of the trajectory point according to the spatial pointwise feature of the trajectory points within a preset observation time interval; and performing motion trajectory prediction on the object according to the spatial pointwise feature and the temporal pointwise feature of the trajectory points. The present disclosure promotes the accuracy of the motion trajectory prediction.