Boundary Line Detection Using Learned Model Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Rule-based approaches for detecting boundary lines in images are influenced by external environmental factors such as camera characteristics and weather conditions, leading to inconsistent performance.
Innovation Solution
A device and method utilizing a learned model, specifically a convolutional neural network (CNN), to detect boundary lines by specifying and grouping estimated points within captured images, reducing environmental influence through machine learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a rule-based approach is used to detect boundary lines by determining pixels based on brightness changes, then the detection process is simple and fast, but the detection accuracy is reduced due to influence from external environment factors such as camera characteristics and weather conditions
Solution Approach 1:
The patent replaces the rule-based mechanical detection system with a machine learning-based system. Instead of using fixed brightness change thresholds and geometric rules, the system employs a trained neural network model that automatically learns optimal detection patterns from training data, substituting manual rule-based processing with adaptive intelligent processing.
Solution Approach 2:
The patent transforms the detection approach by changing from fixed parameter thresholds to dynamic parameter adaptation. The machine learning model learns optimal parameter values during training and adapts them based on input image characteristics, allowing the system to adjust detection parameters dynamically rather than using static brightness thresholds.
2Device complexity
If a rule-based approach is used to detect boundary lines, then the system structure is simple, but the reliability is reduced due to sensitivity to external environmental factors
Solution Approach 1:
The patent replaces the simple but unreliable rule-based system with a machine learning system that maintains structural simplicity through standardized model architecture while achieving enhanced reliability. The neural network model provides consistent detection performance across varying environmental conditions by learning robust feature representations during training.
Solution Approach 2:
The patent applies preliminary action by training the machine learning model in advance with extensive training data that encompasses various environmental conditions. This pre-training phase allows the system to learn and adapt to different camera characteristics, weather conditions, and lighting scenarios before actual deployment, ensuring reliable performance without complex runtime adjustments.
3Reliability
If a learned model is used to specify estimated points for boundary line detection, then the influence of external environment is reduced, but the processing complexity increases
Solution Approach 1:
The patent replaces complex manual rule-based processing with a streamlined machine learning inference process. Although the model training is complex, the actual detection process uses the trained model to directly predict boundary line positions from input images, simplifying the runtime processing while maintaining environmental robustness.
Data Source
AI summary
A CPU 11 of a boundary line detection device 10 according to an embodiment of the present invention is configured to function as an image acquisition component 112 that acquires images captured (inputted) via an image input device (a camera or the like), captured images received via a network, and so forth, and a boundary line detector 114 that detects a boundary line included in these captured images. The boundary line detector 114 is configured to specify, using a learned model 152, a plurality of estimated points estimated to be points on a boundary line included in a captured image and to detect the boundary line included in the captured image on the basis of at least the specified estimated points.