Trajectory Prediction Model Using Virtual Labels and Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face a shortage of training data for learning machine learning models to predict object trajectories, which hinders accurate trajectory prediction.
Innovation Solution
A technique that generates virtual labels associating target images, object positions, and trajectory reliability to learn a state transition model, compensating for the lack of training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning models are trained using existing training data, then trajectory prediction can be performed, but the accuracy is insufficient due to shortage of training data
Solution Approach 1:
The patent generates virtual trajectory labels by copying and transforming existing trajectory data through a trajectory generation unit. This creates synthetic training data that mimics real trajectory patterns, effectively multiplying the available training data without requiring additional physical measurements or observations.
Solution Approach 2:
The system performs preliminary trajectory predictions using an initial machine learning model, then uses these predictions to generate virtual labels before final training. This preliminary action creates a foundation of predicted trajectories that are subsequently refined and used to train a more accurate model.
2Measurement precision
If more training data is collected to improve model accuracy, then trajectory prediction accuracy improves, but data collection time and resources increase
Solution Approach 1:
Instead of collecting additional real-world trajectory data over time, the system copies existing trajectories and generates virtual variations. This synthetic data generation approach achieves the same effect as collecting more data without the time investment required for additional data gathering.
Solution Approach 2:
The system uses its own predicted trajectories as training data through the virtual label generation process. The machine learning model generates predictions that are then fed back into the training pipeline, allowing the system to improve itself without external data sources.
3Quantity of substance
If virtual labels are generated to compensate for training data shortage, then sufficient training data is available, but system complexity increases
Solution Approach 1:
The trajectory generation unit serves multiple functions: it generates virtual trajectories, calculates reliability scores, and creates labeled training data. This multi-functional component reduces the need for separate systems for each task, thereby limiting the increase in overall system complexity.
Solution Approach 2:
The reliability calculation unit acts as an intermediary between trajectory generation and label creation. It provides a quality assessment mechanism that bridges the gap between synthetic trajectory data and usable training labels, managing the complexity of the virtual label generation process.
Data Source
AI summary
An information processing device includes a memory configured to store instructions; and one or more processors configured to execute the instructions to: predict a trajectory of an object included in at least one of a plurality of target images with reference to a plurality of target images; extract a feature of the trajectory and calculate a reliability of the trajectory based on the feature; generate a virtual label in which each of the plurality of target images, a position of the object included in the target image, the trajectory, and the reliability are associated with each other; and learn a state transition model that predicts a state of an object included in a plurality of images by using the virtual label.