3D Feature Prediction From Vehicle Sensor Time Series
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The process of creating training data for deep learning systems in autonomous driving is labor-intensive and inefficient, requiring significant manual effort for data collection and annotation, which limits the accuracy and effectiveness of machine learning models.
Innovation Solution
A method is developed to generate highly accurate training data by using a time series of sensor data from vehicles, including image and odometry data, to create a three-dimensional representation of features like lane lines, which reduces the need for manual annotation and improves data accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual data collection and annotation is used to create training data, then data accuracy can be improved, but labor intensity and time consumption increase significantly
Solution Approach 1:
The system performs preliminary actions by using semi-automatic annotation methods where initial annotations are generated automatically and then refined. This prepares the data in advance with reasonable accuracy, reducing the need for extensive manual review later and decreasing overall time consumption while maintaining acceptable data quality levels.
Solution Approach 2:
The patent introduces an intermediary approach by combining automatic annotation tools with selective manual verification. This intermediary method acts as a bridge between fully automatic and fully manual processes, using automated systems to handle routine tasks and human annotators to focus on complex or ambiguous cases, thereby balancing accuracy and efficiency.
2Measurement precision
If manual annotation is used to label training data features, then labeling accuracy improves, but the effort and cost required increase
Solution Approach 1:
The annotation process is segmented into different levels of complexity. Simple, unambiguous features are annotated automatically or with minimal human intervention, while complex or ambiguous features receive focused manual attention. This segmentation allows the system to allocate human effort efficiently, maintaining high labeling accuracy for critical features while reducing overall manual effort requirements.
Solution Approach 2:
The system implements self-service mechanisms where the annotation tool provides automated suggestions, consistency checks, and quality validation. This allows the annotation process to partially serve itself by automatically handling routine validation tasks and providing guidance to annotators, reducing the overall effort and cost while maintaining labeling accuracy.
3Reliability
If comprehensive training data is collected to improve model performance, then model accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent extracts and focuses on the most critical features and data elements that have the greatest impact on model performance. Rather than processing all available data comprehensively, the system identifies and extracts key training signals, reducing data processing complexity while maintaining model performance by concentrating computational resources on the most informative data subsets.
4Reliability
If accurate three-dimensional feature prediction is achieved, then autonomous driving safety improves, but computational resources required increase
Solution Approach 1:
The system applies partial action by focusing computational resources on predicting only the most critical three-dimensional features that directly impact safety, such as lane line positions and distances to obstacles. Rather than computing all possible spatial features with equal detail, the system allocates computational power selectively to high-priority features, maintaining safety while reducing overall computational resource consumption.
Data Source
AI summary
A processor coupled to memory is configured to receive image data based on an image captured by a camera of a vehicle. The image data is used as a basis of an input to a trained machine learning model trained to predict a three-dimensional trajectory of a machine learning feature. The three-dimensional trajectory of the machine learning feature is provided for automatically controlling the vehicle.