Dynamic Sensor Parameter Optimization for Object Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object recognition systems using sensors like cameras and lidar struggle to achieve optimal results due to fixed or deterministic sensor parameters, limiting the learning ability and efficiency of deep learning algorithms, especially in applications requiring precise measurements like the automotive sector.

Innovation Solution

A procedure and system that dynamically adjust sensor parameters of two sensors (such as camera and lidar) using a Reinforcement Learning algorithm, iteratively refining the sensor settings to maximize object recognition accuracy within specific sensor parameter ranges.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If fixed or deterministic sensor parameters are used, then device complexity is reduced and ease of operation is improved, but object recognition accuracy and algorithm performance are limited

Engineering Contradiction:
Improveobject recognition accuracyVSAvoidsensor parameter adjustment complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent implements dynamic sensor parameters that can be adjusted in real-time based on environmental conditions and object characteristics. The sensor system transitions from fixed parameters to dynamically adaptable parameters, allowing the object recognition algorithm to optimize performance for different scenarios while maintaining manageable complexity through automated control.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent systematically varies sensor parameters such as exposure time, gain, and sampling frequency to find optimal settings for different recognition tasks. By implementing parameter sweeps and adaptive tuning mechanisms, the system explores the parameter space to identify configurations that maximize measurement precision without requiring manual intervention.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If sensor parameters are manually optimized for each scenario, then object recognition accuracy improves, but time consumption and operational complexity increase

Engineering Contradiction:
Improveobject detection precisionVSAvoidparameter optimization time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements pre-training and pre-optimization phases where sensor parameters are optimized offline for various scenarios. These pre-computed parameter sets are stored and automatically selected during operation, eliminating the need for real-time manual optimization while maintaining high detection precision across different conditions.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system employs self-adjusting sensor parameters that automatically adapt to changing conditions without external intervention. The object recognition algorithm provides feedback to the sensor control system, which autonomously tunes parameters to maintain optimal performance, reducing both time loss and operational complexity.

Inventive Principle:
Principle #25Self-service

3Productivity

If deep learning algorithms are used with fixed sensor parameters, then implementation simplicity is maintained, but the full potential of the algorithms cannot be utilized due to occlusion and unfavorable parameters

Engineering Contradiction:
Improvealgorithm efficiencyVSAvoidobject interpretation reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements a closed-loop system where the object recognition algorithm provides feedback on detection quality and confidence levels. This feedback is used to dynamically adjust sensor parameters, creating a responsive system that adapts to challenging conditions such as occlusion. The feedback mechanism enables the system to recover from suboptimal detections by adjusting parameters to capture additional informative data.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3945452B1Method for optimising sensor parameters
Publication Date: 2025.04.09 DEUTSCHE TELEKOM AG
  • EP3945452B1 patent drawingFigure 1
  • EP3945452B1 patent drawingFigure 2
  • EP3945452B1 patent drawingFigure 3~4

AI summary

The present invention relates to a method for adjusting the sensor parameters of at least one sensor (11) when using the at least one sensor (11) in functional connection with an object recognition algorithm (20) during object recognition, wherein the sensor parameters (16) to be adjusted are set on the at least one sensor (11), the at least one sensor (11) is activated with the set sensor parameters (15), and the at least one sensor (11) sensorially acquires data (12) and provides it to the object recognition algorithm (20), the object recognition algorithm (20) determines a result (21), comprising at least one detected object (13) with a match value assigned to the detected object, and provides it to a machine learning algorithm (30).The machine learning algorithm (30) varies the respective sensor parameters of the at least one sensor (11) within a respective sensor parameter range and provides the respective varied sensor parameters as adjustable sensor parameters (16) for the at least one sensor (11), and steps a. to d. are iteratively repeated until the respective match value assigned to the at least one detected object (13) reaches a maximum within at least one of the respective sensor parameter ranges. Furthermore, the present invention relates to a computer-readable medium for adjusting adjustable sensor parameters of at least one sensor.