ADS Sensor-View Label Transformation for Drivable Space Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The lack of appropriate training data for neural networks in autonomous and assisted driving systems limits their effectiveness, particularly when environmental sensors have different views of the traffic environment than the available sample data.
Innovation Solution
A method that transforms top view or polar representations of traffic environments into perspective representations aligned with the view of environmental sensors, using a transformation matrix that includes a sensor calibration matrix, allowing the use of diverse sensor data for training neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If training data is collected from sensors with different viewing perspectives (e.g., top view vs. perspective view), then the quantity of training data increases, but the data becomes unsuitable for training neural networks requiring consistent coordinate systems
Solution Approach 1:
A transformation matrix serves as an intermediary tool to convert coordinate systems between different sensor perspectives. The matrix enables data from top-view sensors to be transformed into perspective-view coordinates, making heterogeneous sensor data compatible for neural network training while preserving the quantity benefits of multi-sensor data collection
Solution Approach 2:
The patent changes the coordinate system parameters of training data through mathematical transformation. By applying the transformation matrix, data parameters (coordinates) are converted from one reference frame to another, enabling seamless integration of data from sensors with different viewing perspectives without losing data quantity advantages
2Adaptability or versatility
If training data is transformed between different coordinate systems using transformation matrices, then the adaptability of training data to different sensors improves, but the processing complexity and time increase
Solution Approach 1:
The transformation matrix is pre-calibrated based on the geometric relationship between sensors before actual training data processing. This preliminary setup stores the transformation parameters, allowing rapid coordinate conversion during training without performing complex real-time calculations, thus reducing processing time while maintaining high adaptability
3Adaptability or versatility
If multiple types of sensors (camera, radar, lidar) are used to collect training data, then the diversity and quantity of training data improve, but the difficulty of processing and integrating different sensor formats increases
Solution Approach 1:
The transformation matrix framework provides a universal processing mechanism that handles multiple sensor types (camera, radar, lidar) through a unified coordinate conversion approach. This multi-functional system accommodates diverse sensor data formats by applying appropriate transformation matrices for each sensor type, reducing the need for separate processing pipelines and lowering overall system complexity
Data Source
Figure 1~2a
Figure 2b~3
Figure 4a~4b
AI summary
The present disclosure relates to a vehicle, an autonomous/assisted driving system (ADS) a computer program, an apparatus (600), and a method (100) for an autonomous/assisted driving system. The method (100) comprises obtaining (110) a top view representation of labels of a traffic environment in Cartesian coordinates from sample data. The method (100) also comprises obtaining (120) a transformation matrix for transforming Cartesian coordinates in an observation coordinate system of a perspective of an environmental sensor of the ADS. Further, the method (100) provides for applying (130) the transformation matrix to the top view representation of the labels to obtain a perspective representation of the traffic environment in the observation coordinate system.