Object Centering Using Depth Segmentation for Detection Errors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object centering technologies face challenges in accurately detecting and centering diverse object types due to the difficulty in collecting and training data for complex object types, leading to errors in object detection results that affect the centering process.
Innovation Solution
An object centering method and device that perform foreground object segmentation using depth and face images, determine a relation with a threshold, and execute an object centering process based on the determination result to generate an object centering image, incorporating techniques like bilinear interpolation and progressive centering to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If object detection is used for diverse object types, then the system can process various objects, but the detection accuracy deteriorates due to difficulty in collecting and training data for complex object types
Solution Approach 1:
The patent segments the object detection task into two parts: using a pre-trained detection model for general object detection, and using a dedicated foreground object segmentation module for accurate foreground object identification. This segmentation allows the system to maintain high accuracy for foreground objects while still supporting diverse object types through the pre-trained model's broad knowledge.
Solution Approach 2:
The patent introduces a foreground object segmentation module as an intermediary between the pre-trained detection model and the final centering process. This intermediary component refines the detection results by specifically identifying foreground objects, thereby improving detection accuracy for diverse object types without requiring retraining the entire system.
2Adaptability or versatility
If pre-trained object detection models are used, then the system can process various object types, but errors in detection results occur due to insufficient training data for complex object types
Solution Approach 1:
The patent performs preliminary foreground object segmentation before the centering process. By pre-identifying and separating foreground objects from background using the segmentation module, the system prepares reliable detection results in advance, ensuring that subsequent centering operations are based on accurate object identification regardless of object type complexity.
Solution Approach 2:
The patent implements a feedback mechanism where the foreground object segmentation results are used to correct and refine the initial detection results. The system continuously iterates between detection and segmentation, using segmentation feedback to improve detection accuracy for complex object types while maintaining versatility.
3Measurement precision
If foreground object segmentation is performed using depth and face images, then the centering accuracy is improved, but the processing complexity increases
Solution Approach 1:
The patent designs the foreground object segmentation module to handle multiple object types and scenarios universally. By using a multi-functional approach that combines depth image processing, face image analysis, and foreground segmentation, the system achieves high centering accuracy across diverse applications without requiring separate specialized processing for each object type, thereby managing complexity through generalization.
Data Source
AI summary
An object centering method is disclosed herein. A processor reads at least command stored in a memory and executes the object centering method. The object centering method includes following steps: performing a foreground object segmentation process according to a depth image and a face image to generate a foreground object image; determining a relation between the foreground object image and a predetermined threshold to generate a determination result; and performing an object centering process to the foreground object image according to the determination result to generate an object centering image, or performing the object centering process to the face image according to the determination result to generate the object centering image.