Vision Training Data Fusion for Moving Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for training vision-based machine learning models for autonomous vehicles rely heavily on human-labeled data, which is inefficient, limited in scale, and prone to errors, particularly in generating accurate distance and velocity data for objects, hindering the development of reliable autonomous driving systems.
Innovation Solution
A network service processes combined data from vision systems and additional sensors like radar and LIDAR to generate updated training datasets, leveraging ground truth data from radar systems to refine vision-only machine learning models, enabling rapid data gathering and transmission for continuous model updates, thus reducing sensor hardware complexity while enhancing accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If human-labeled data is used for training vision-based machine learning models, then the training process is simple to implement, but the data gathering speed is slow and the scale is limited
Solution Approach 1:
The patent introduces an intermediary processing system that combines vision data with ground truth data from other sensors (radar, LIDAR) to generate training data. This intermediary layer allows the system to leverage multiple data sources without requiring all sensors to be present in every vehicle, thus increasing data gathering speed while managing hardware complexity through selective sensor deployment.
Solution Approach 2:
The patent creates synthetic training data by copying and combining information from multiple sources. Specifically, it uses ground truth data from radar and LIDAR systems to create accurate distance and velocity labels that are then paired with vision data, effectively copying the reliability of multi-sensor systems into training datasets that can be used by vision-only models.
2Measurement precision
If vision-only machine learning models are used, then the sensor hardware complexity is reduced, but the accuracy of distance and velocity detection deteriorates
Solution Approach 1:
The patent performs preliminary actions by pre-processing vision data and combining it with ground truth data from other sensors before training the model. This preliminary combination of data sources during the training phase allows the vision-only model to learn from enriched data that contains accurate distance and velocity information, compensating for the lack of dedicated distance/velocity sensors in deployment.
Solution Approach 2:
The patent changes the parameters of the training data by incorporating additional dimensions (distance, velocity) derived from ground truth sensors into the vision data framework. This parameter enrichment transforms the training dataset from simple image-label pairs to multi-dimensional data structures that teach the vision model to infer distance and velocity accurately without requiring physical distance/velocity sensors.
3Reliability
If multiple sensors (radar, LIDAR) are integrated to improve detection accuracy, then the measurement precision improves, but the device complexity increases
Solution Approach 1:
The patent segments the sensor system into two distinct groups: vehicles equipped with multiple sensors (radar, LIDAR, vision) that collect and transmit ground truth data, and vision-only vehicles that receive and utilize processed training data. This segmentation allows the system to achieve high reliability through multi-sensor data collection while managing complexity by allowing vision-only deployment for the actual autonomous driving function.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach significantly improves the speed and scale of training data creation, leading to more accurate machine learning models for autonomous vehicles by utilizing diverse sensor inputs to supplement and validate vision data, resulting in enhanced object detection and classification capabilities.
Implementation Method 1
A network service processes combined data from vision systems and additional sensors like radar and LIDAR to generate updated training datasets, leveraging ground truth data from radar systems
Implementation Method 2
A network service processes combined data from vision systems and additional sensors like radar and LIDAR to generate updated training datasets, leveraging ground truth data from radar systems
Data Source
AI summary
Systems and methods for training machine learning models utilized for autonomous driving. An example method includes obtaining a set of data corresponding to the operation of a vehicle, wherein the set of data includes a first set of data corresponding to the operation of a vision-based detection system and a second set of data corresponding to the operation of a non-vision-based detection system, wherein the first and second set of data corresponding to a common timestamp; processing the first set of data to correspond to a common format for detection; processing the second set of data to correspond to the common format for detection; combining the processed first set of data and the processed second set of data to form a common set of data; processing the combined set of data; and training a machine learning model for vision-based detection system based on the processing combined set of data.