Adaptive Pyramid Image Generation for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection technologies face challenges in balancing accuracy and speed, particularly in detecting targets like faces, eyes, or irises across varying sizes and locations within images, especially when the target's size changes rapidly or is distant from the camera.
Innovation Solution
The method involves adaptively generating a pyramid image based on information from previous frames, such as pyramid ID, location, and size of the target object, and setting a scan area to efficiently detect the target object from the pyramid image, reducing the time required for image generation and detection while maintaining high detection probability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full-range pyramid image is generated for object detection, then detection accuracy is improved, but detection speed deteriorates due to processing large numbers of sub-images
Solution Approach 1:
The patent applies local quality by generating pyramid images with different ranges (full-range, first-range, second-range) based on the detection needs and target characteristics. Instead of uniformly processing all sub-images, the system selectively generates pyramid images of appropriate ranges, concentrating computational resources on relevant areas and scales, thereby improving detection speed without sacrificing accuracy for the specific detection task
Solution Approach 2:
The patent implements dynamics by adaptively adjusting the pyramid image generation range based on information from previous frames, such as target location and size. The system dynamically switches between full-range, first-range, and second-range pyramid images depending on whether the target is stationary or moving, and whether it occupies a large or small area, optimizing the balance between detection accuracy and speed in real-time
2Productivity
If pyramid image generation range is reduced to improve speed, then detection speed is improved, but detection accuracy deteriorates
Solution Approach 1:
The system dynamically adjusts the pyramid image generation range based on target characteristics and detection requirements. When the target is stationary and occupies a large area, full-range pyramid images are generated for comprehensive detection. When the target is moving or occupies a small area, first-range or second-range pyramid images are generated to improve speed while maintaining sufficient detection accuracy
Solution Approach 2:
The patent applies local quality by matching the pyramid image generation range to the specific detection scenario. For distant or small targets, first-range or second-range pyramid images focus computational resources on relevant scales. For close or large targets, full-range pyramid images provide comprehensive coverage, ensuring detection accuracy is maintained for each specific case
3Measurement precision
If pyramid images of multiple ranges are generated to handle varying target sizes, then detection accuracy is improved, but processing time increases
Solution Approach 1:
The system dynamically selects which pyramid image ranges to generate based on information from previous frames, such as target location, size, and motion status. Instead of always generating all three ranges (full, first, second), the system adaptively chooses the appropriate range, reducing unnecessary processing time while maintaining detection accuracy for varying target sizes
Solution Approach 2:
The patent applies local quality by generating pyramid images of specific ranges matched to the detected target characteristics. For small or distant targets, first-range or second-range pyramid images are generated to focus on relevant scales. For large or close targets, full-range pyramid images are generated, optimizing processing time by avoiding unnecessary generation of all ranges for every detection case
Data Source
AI summary
An object detection method and apparatus are provided. The object detection method may include adaptively generating a pyramid image corresponding to a current frame based on information associated with a target object detected from a previous frame.