Object Detection Apparatus Using Likelihood-Based Region Clipping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for detecting objects from images are inefficient when applied to moving images, as they require processing all regions uniformly, leading to suboptimal detection efficiency, especially for objects entering or exiting the frame.
Innovation Solution
An image processing apparatus that uses object likelihood information stored for each frame to determine the most likely positions for object detection, shifting the search focus based on previous frames' data and motion vectors, thereby reducing unnecessary processing and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all regions in the image are processed uniformly to detect objects, then detection coverage is ensured, but detection efficiency deteriorates due to unnecessary processing of regions without objects
Solution Approach 1:
The patent applies local quality by differentiating processing strategies for different regions of the image. Regions with high object likelihood (determined from previous frames and motion vectors) undergo full detection processing, while regions with low likelihood use simplified or skipped processing, optimizing the balance between coverage and efficiency
Solution Approach 2:
The patent performs preliminary action by calculating object likelihood information from previous frames and motion vectors before executing the full detection process. This preliminary assessment allows the system to pre-identify regions worthy of full processing, avoiding unnecessary computation in regions unlikely to contain objects
2Productivity
If the search region is restricted based on previous detection results, then processing load is reduced, but detection accuracy deteriorates when objects enter or exit the frame
Solution Approach 1:
The patent applies dynamics by making the search region adaptive rather than static. The search region automatically expands to cover areas with high object likelihood calculated from motion vectors and previous frame data, allowing the system to dynamically adjust the processing scope based on predicted object locations, thus maintaining accuracy while reducing unnecessary processing
Solution Approach 2:
The patent uses feedback from previous detection results and motion analysis to continuously update the search region. The object likelihood information from previous frames feeds into the current frame's processing decisions, creating a closed-loop system that adapts to object movement patterns while maintaining detection accuracy
Data Source
AI summary
An image processing apparatus includes a moving image input unit configured to input a moving image, an object likelihood information storage unit configured to store object likelihood information in association with a corresponding position in an image for each object size in each frame included in the moving image, a determination unit configured to determine a pattern clipping position where a pattern is clipped out based on the object likelihood information stored in the object likelihood information storage unit, and an object detection unit configured to detect an object in an image based on the object likelihood information of the pattern clipped out at the pattern clipping position determined by the determination unit.