Image Segmentation for Robust Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image tracking techniques fail to accurately track objects with complex geometrical shapes or those that change shape between images, due to their reliance on predetermined feature extraction areas, leading to inaccuracies and increased processing load.
Innovation Solution
An image processing device that segments images based on pixel value similarity, sets a reference area dynamically, extracts feature quantities from these areas, and uses them to track objects by determining corresponding segments within an area of interest, allowing for robust tracking even with shape changes and complex geometries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional techniques use predetermined feature extraction areas, then the tracking process is simple, but tracking accuracy deteriorates for objects with complex geometrical shapes or shape changes
Solution Approach 1:
The patent divides the image into multiple segments based on pixel value similarity, creating a segmented map that adapts to the actual object shape. This segmentation approach allows the system to handle complex geometrical shapes by treating each segment independently, thereby improving tracking accuracy without requiring a predetermined fixed-area extraction method
Solution Approach 2:
The patent dynamically adjusts the feature extraction area by identifying the actual object boundaries through segmentation and using the largest connected component or bounding box around it. This dynamic adaptation allows the system to accommodate shape changes between frames, improving tracking accuracy for objects that deform or change geometry while maintaining reasonable processing complexity
2Reliability
If conventional techniques use fixed feature extraction areas, then processing is faster, but reliability deteriorates when objects change shape between images
Solution Approach 1:
The patent performs segmentation and identifies the object boundary in advance before feature extraction. By pre-processing the image to determine the actual object extent and creating a mask or bounding box, the system prepares the correct extraction area beforehand, ensuring reliable tracking even when shapes change. This preliminary action prevents incorrect feature extraction that would occur with fixed areas, while the efficient segmentation algorithm maintains acceptable processing speed
3Measurement precision
If conventional techniques extract features from predetermined areas, then the system is simpler to implement, but measurement precision worsens for objects with complex geometries
Solution Approach 1:
The patent implements segmentation by dividing the image into regions of similar pixel values and identifying the largest connected component or bounding box around the object. This segmentation-based approach accurately captures complex geometrical shapes by following actual pixel boundaries rather than using predetermined fixed areas, significantly improving feature extraction accuracy while adding manageable computational steps
Solution Approach 2:
The patent extracts only the relevant portion of the image by identifying the largest connected component or creating a bounding box around the segmented object. This extraction of the precise object region eliminates unnecessary background or unrelated areas from feature extraction, improving measurement precision by focusing computational resources only on the actual object pixels
Data Source
AI summary
An image processing device includes: a reference area setting unit which sets a reference area including an indicated segment; an extraction unit which extracts an interest object feature quantity indicating a first feature from the reference area; an interest area setting unit which sets an area of interest in a third image, based on a relationship between a position of a feature point extracted from a feature area which is an area corresponding to an object of interest and included in a second image and a position of a feature point in the third image corresponding to the extracted feature point; and a tracking unit which determines for each of two or more of plural segments included in the area of interest with use of the interest object feature quantity whether the segment is a segment corresponding to the object of interest.