Object Part Detection Model Correction via Reliability-Based Repositioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques using deformable part models struggle to accurately detect the position of object parts, especially when there are significant changes in the overall shape or relative position of partial regions within the same category, such as a person performing various movements, as they fail to sufficiently select reliable partial regions from learned images.
Innovation Solution
An image processing apparatus that includes an acquisition unit for acquiring images, a first detection unit for detecting candidate regions using a learned model, a selection unit for identifying high-reliability and low-reliability parts, and a correction unit that adjusts the model by repositioning low-reliability parts based on high-reliability parts, thereby improving detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a deformable part model is used to detect object parts, then the detection can be performed on objects with varying shapes and attitudes, but the detection precision deteriorates when there are large changes in relative position or direction of partial regions
Solution Approach 1:
The object detection model is segmented into multiple parts (root part and plurality of parts) with hierarchical relationships. Each part is detected and positioned independently, allowing the system to handle variations in object shape and attitude while maintaining precise detection of individual part positions through separate candidate region generation and selection processes.
2Manufacturing precision
If partial regions are selected from learned images using gradient average methods, then the technique works well for objects with fixed shapes, but it fails when objects have large changes in overall shape or relative position of partial regions
Solution Approach 1:
The system dynamically adjusts part positions and attitudes based on detected candidate regions rather than using fixed partial regions from learned images. The model allows parts to vary in position and orientation according to the actual object姿态, enabling adaptation to objects with large shape variations while maintaining detection accuracy through dynamic position adjustment.
Solution Approach 2:
The system changes the parameters of partial regions (position, size, orientation) based on detected candidate regions from the image. Instead of using fixed parameters from learned images, the system adjusts these parameters dynamically to match the actual object configuration, enabling effective detection of objects with varying attitudes and shapes.
3Measurement precision
If the model is corrected by repositioning low-reliability parts based on high-reliability parts, then the detection accuracy of object parts is improved, but the processing complexity increases
Solution Approach 1:
The system uses a feedback mechanism where high-reliability parts serve as reference points to correct and reposition low-reliability parts. The detection results of high-reliability parts feed back into the model correction process, adjusting the positions of low-reliability parts to improve overall detection accuracy while maintaining a structured correction workflow.
Data Source
AI summary
An image processing apparatus includes an acquisition unit, a first detection unit, a selection unit, and a correction unit. The acquisition unit acquires an image including a target object having a plurality of parts. The first detection unit detects a candidate region of each of the plurality of parts of the target object included in the acquired image using a previously learned model. The selection unit selects, based on the candidate region detected by the first detection unit, a first part having relatively high reliability and a second part having relatively low reliability from among the plurality of parts. The correction unit corrects the model by changing a position of the second part based on the first part selected by the selection unit.