Relative Bearing Prediction for Dense Target-Time Sensor Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor technologies produce sparse point clouds with limited information density, which hampers the performance of AI modules and real-time object detection and tracking, particularly in autonomous driving applications.
Innovation Solution
A method that predicts sensor data by using an object motion model to virtually shift measurements from preliminary times to a target time, combining actual and predicted relative bearings to create a denser point cloud, enhancing data density for improved processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If sensor measurements are taken at discrete time points, then the sensor can capture relative bearing data, but the resulting point cloud becomes sparse with limited information density
Solution Approach 1:
The system performs preliminary measurements at multiple time points before the target time, then uses an object motion model to predict what the measurement would have been at the target time. This preliminary action of measuring and predicting in advance allows the system to compensate for the sparsity caused by discrete time sampling, effectively increasing information density without requiring continuous measurements.
Solution Approach 2:
The system creates a virtual copy of the preliminary measurement by predicting the relative bearing at the target time based on the object motion model. This predicted measurement is then combined with the actual target time measurement, effectively duplicating the information content and increasing the density of the point cloud without requiring additional physical measurements.
2Quantity of substance
If multiple preliminary measurements are taken before the target time, then more data points are available for processing, but the computational complexity increases due to prediction requirements
Solution Approach 1:
The system changes the temporal parameter of the measurements by predicting preliminary measurements to the target time using an object motion model. This parameter transformation allows multiple measurements to be aligned to a common time reference, increasing the effective number of measurement points while managing computational complexity through efficient motion model-based predictions rather than requiring full reprocessing of all raw data.
3Speed
If the sensor operates in real-time with discrete measurements, then the system responds quickly to changes, but the point cloud remains sparse which hampers AI module performance
Solution Approach 1:
The system performs predictions of measurement values at the target time based on preliminary measurements and object motion models. This preliminary action allows the system to prepare denser point cloud data in advance, improving AI module performance without sacrificing real-time response capability, as the predictions are computed efficiently based on the motion model rather than requiring additional measurement time.
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
A method for producing sensor data (10) involves a sensor (2) detecting a relative bearing (R) of an object (6) at a sensor null point (4) as a preliminary sensor bearing (VSL1,2,...) at preliminary times (t1,2,...), and a target sensor bearing (ZSL) at a target time (tz), and a computing module (12) predicting for the preliminary sensor bearings (VSL1,2,...), on the basis of an object motion model (14), a respective prediction sensor bearing (PSL1,2,...) at the target time (tz), and the sensor data (10) being formed as a set comprising the target sensor bearing (ZSL) and the prediction sensor bearing (PSL1,2,...) at the target time (tz). A method for controlling a motor vehicle (40) in surroundings (44) having the object (6) involves the sensor data (10) with regard to the object (6) being produced and an actuator (42) being triggered with regard to the object (6) on the basis of the sensor data (10). A sensor module (50) contains the sensor (2) and the computing module (12) and an output (52) for the sensor data (10). A motor vehicle (40) contains the sensor module (50).