Dual-Model Image Recognition with Overlap Invalidation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition techniques struggle with accurate detection of objects, such as pedestrians, in images taken by vehicles, as a part of a detection target in a nearby area may be detected as a target by a recognition dictionary for a distant area, leading to improper detection.
Innovation Solution
An image recognition device and method that utilize two detection models trained by machine learning: one for nearby areas with larger image sizes and another for distant areas with smaller image sizes. The system detects regions in the captured image using these models and includes a determination unit that invalidates the detection of overlapping regions to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single recognition dictionary is used for both near and far areas, then the device complexity is reduced, but the detection precision deteriorates because parts of nearby objects may be incorrectly detected as distant objects
Solution Approach 1:
The patent divides the single recognition dictionary into multiple area-specific dictionaries (near area dictionary and far area dictionary). Each dictionary is trained on images from its specific distance range, allowing the system to select the appropriate dictionary based on the detected object's distance, thereby improving detection precision while maintaining manageable complexity through modular organization
Solution Approach 2:
The patent applies different recognition characteristics to different spatial zones by creating area-specific recognition dictionaries. The near area dictionary uses features optimized for close-range objects while the far area dictionary uses features optimized for distant objects, allowing each region to be processed with locally-appropriate detection parameters
2Measurement precision
If multiple recognition dictionaries for different areas are used, then the detection precision is improved, but the device complexity increases due to managing multiple dictionaries
Solution Approach 1:
The patent implements a dynamic dictionary selection mechanism that automatically chooses the appropriate recognition dictionary based on the detected object's distance from the camera. The system adjusts which dictionary to use in real-time based on spatial information, making the multi-dictionary system adaptive rather than static, thereby managing complexity through intelligent routing
Solution Approach 2:
The patent introduces a distance determination unit as an intermediary that bridges the image acquisition unit and the recognition process. This intermediary calculates the distance to detected objects and uses that information to select the appropriate recognition dictionary, thereby managing the complexity of multiple dictionaries through a centralized selection mechanism
3Measurement precision
If machine learning models with large image sizes are used for nearby objects, then the detection precision for nearby objects is improved, but the processing time and computational resources increase
Solution Approach 1:
The patent segments the image processing task by distance, applying different model configurations to different spatial zones. Nearby objects are processed with high-resolution models that require more computational resources, while distant objects are processed with lower-resolution models, thereby optimizing the balance between detection precision and processing time for each spatial region
Solution Approach 2:
The patent changes the input image size parameter based on the detection task requirements. For nearby objects, larger image sizes are used to capture fine details, while for distant objects, smaller image sizes are sufficient, thereby adjusting computational requirements to match the actual information needs of each detection scenario
Data Source
AI summary
An image recognition device includes: an image acquisition unit that acquires a captured image; a first detection unit that detects a first region including a detection target in the captured image using a first detection model trained by machine learning with an image having an image size of a predetermined value or more as input; a second detection unit that detects a second region including the detection target in the captured image using a second detection model trained by machine learning with an image having an image size of less than the predetermined value as input; and a determination unit that invalidates detection of either one of the first region and the second region when the first region and the second region overlap in the captured image.