Vehicle Driving Data Labeling for Autonomous Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The collection of training data for neural networks in autonomous driving predominantly relies on manual labor, requiring substantial human resources and time with low efficiency, leading to inadequate comprehensiveness and accuracy, which affects the training efficiency and adaptability of the neural network.
Innovation Solution
A method for automatically collecting training data using vehicle-mounted devices that integrate sensors and navigation components to gather driving data, generate vehicle trajectories, and label data, enabling efficient generation of training data without manual intervention, and a server-based training process to refine the neural network model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual labor is used to collect training data, then data collection can be performed with simple equipment, but the efficiency and productivity of data collection is low
Solution Approach 1:
The system enables autonomous data collection where the vehicle-mounted device automatically collects driving data, generates trajectories, and labels data without human intervention. The neural network model autonomously processes and labels the collected data, eliminating the need for manual annotation while maintaining high productivity.
Solution Approach 2:
The patent replaces manual mechanical data annotation with automated computational processing. The neural network model substitutes human labor in the data labeling process, transforming a manual task into an automated computational task that significantly improves efficiency.
2Measurement precision
If manual labor is used to collect training data, then the process is simple to operate, but the comprehensiveness and accuracy of the collected data is insufficient
Solution Approach 1:
The system continuously collects driving data during normal vehicle operation without interruption. Data collection occurs continuously as the vehicle operates, and the neural network continuously processes and labels the data, eliminating the stop-start nature of manual data collection and improving both accuracy and time efficiency.
Solution Approach 2:
The system implements feedback mechanisms where the neural network model is trained on collected data, then used to label new data, which is fed back into the training process. This iterative feedback loop continuously improves data accuracy and model performance over time.
3Productivity
If manual labor is used for data collection, then resource requirements are low, but the training efficiency of the neural network is affected
Solution Approach 1:
The system uses the vehicle's own computational resources and collected data to train and improve its neural network model autonomously. The model serves itself by automatically labeling data and improving its own training dataset without requiring external human annotation resources.
Solution Approach 2:
The vehicle-mounted device performs multiple functions: it collects driving data, generates trajectories, labels data using the neural network, and trains the model. This multi-functional system eliminates the need for separate manual data annotation processes, improving training efficiency while reducing human resource requirements.
Data Source
AI summary
A method for obtaining data are provided. The method for obtaining data includes collecting first driving data of a vehicle at first preset time, and collecting a plurality of data sets of second driving data of the vehicle within a first preset period. The vehicle driving trajectory is generated according to the plurality of data sets of the second driving data, and training data is generated according to the first driving data and the vehicle driving trajectory. These method can improve the efficiency of collecting training data and the efficiency of training the neural network.