Image Processing Apparatus for High-Speed Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing techniques for object detection, such as face and human body detection in monitoring camera images, face challenges in achieving high-speed detection when objects of varying sizes are present in the same frame, as they require using collation patterns for both small and large objects, leading to inefficient processing.
Innovation Solution
An image processing apparatus and method that acquires an image and detects patterns of different sizes from distinct regions within the image, using a recursive reduction process to generate multiple image layers for pattern matching, allowing for efficient detection of objects regardless of size by optimizing the detection process through a correspondence table and layer settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If collation patterns for both small and large objects are used to detect all objects in the image, then detection completeness is improved, but processing speed deteriorates
Solution Approach 1:
The image is divided into multiple detection regions (first detection region, second detection region, etc.) based on depth information. Each region is assigned specific collation patterns corresponding to object sizes appropriate for that depth. This segmentation allows the system to use only necessary patterns for each region, maintaining detection completeness while reducing overall processing load and improving speed.
2Measurement precision
If collation patterns corresponding to small and large objects are used, then detection accuracy for varying object sizes is improved, but computational complexity increases
Solution Approach 1:
Different collation patterns are assigned to different detection regions based on their depth characteristics. The first detection region uses patterns for a first object size, while the second detection region uses patterns for a second object size. This local optimization ensures each region uses patterns matched to its specific conditions, maintaining high detection accuracy while reducing the total number of pattern matches required across the entire image.
Data Source
AI summary
An image corresponding to a pattern having a first size is detected from a first detection region in an acquired, first image, and an image corresponding to a pattern having a second size is detected from a second detection region different from the first detection region in the first image.