Vehicle Object Recognition With Self-Correcting Multisensor Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object recognition systems in automated driving and driver assistance require extensive annotated data for training machine learning algorithms, particularly for estimating indirectly measurable variables like velocity and acceleration, which is costly and time-consuming, and often relies on a single type of sensor, limiting accuracy and necessitating complex data labeling.
Innovation Solution
A self-monitoring learning method using a multisensor state estimation approach, where predicted object states are compared to actual measurements to correct the machine learning algorithm, allowing for the estimation of object states without annotated data and utilizing multiple sensor types for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are trained using annotated sensor data to estimate indirectly measurable variables such as velocity and acceleration, then the accuracy of object state estimation is improved, but the time and cost expenditure for data labeling increases significantly
Solution Approach 1:
The system performs self-training by using its own predictions as training data. The machine learning algorithm generates predictions from sensor data, these predictions are then used to train the algorithm without requiring external annotated data. This self-service mechanism eliminates the time-consuming manual labeling process while maintaining estimation accuracy through continuous self-improvement
Solution Approach 2:
The system implements a feedback loop where predictions are continuously compared with actual sensor measurements, and the deviations are used to correct and retrain the machine learning algorithm. This feedback mechanism allows the system to automatically improve its accuracy over time without external annotated data, resolving the contradiction between maintaining high precision and avoiding time-consuming labeling
2Measurement precision
If multiple sensor types are utilized for object recognition, then the accuracy and robustness of estimation is improved, but the complexity of data integration and processing increases
Solution Approach 1:
The machine learning algorithm is designed to process multiple sensor types (lidar, radar, cameras) through a unified framework. The same algorithmic structure handles different sensor data formats and characteristics, making the system multi-functional without requiring separate processing pipelines for each sensor type. This universality reduces integration complexity while maintaining the accuracy benefits of multisensor fusion
Solution Approach 2:
The system transforms measurements from different sensor types into a common state space representation that the machine learning algorithm can process uniformly. By changing the parameters and coordinate systems of different sensors to a standardized format, the system simplifies data integration while preserving the complementary information from multiple sensor types, thus improving accuracy without proportionally increasing complexity
3Measurement precision
If lidar sensors are used for indirect measurement of dynamic properties, then the measurement accuracy is improved, but the system becomes restricted to a single sensor type requiring complex data labeling
Solution Approach 1:
The machine learning algorithm creates a universal processing framework that can handle multiple sensor types (lidar, radar, cameras) with a single unified approach. This universality allows the system to achieve lidar-level accuracy for dynamic property estimation while being adaptable to various sensor types, eliminating the restriction to single-sensor systems and reducing the need for complex type-specific labeling
Data Source
AI summary
A method recognizes objects in an environment of a vehicle including a sensor system having at least one sensor unit for registering the environment and an evaluation unit for evaluating sensor data provided by the at least one sensor unit. The method includes receiving the sensor data in the evaluation unit, the sensor data including a plurality of chronologically successive measurements; inputting the sensor data into a machine learning module; and outputting an object state of at least one object estimated based on the sensor data by way of the machine learning module. The method further includes determining a plurality of chronologically successive future object states based on the estimated object state; ascertaining deviations between the future object states and measurements chronologically corresponding to the future object states based on the sensor data; and correcting a machine learning algorithm of the module based on the deviations.