Depth Image Object Tracking via 2D-3D Segmented Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Connected-component labeling techniques face performance issues and increased processing time when applied to complex three-dimensional depth images, leading to slow operation and excessive battery consumption.
Innovation Solution
Perform connected-component labeling in both screen dimensions and depth dimensions, using contours to identify objects, and implement techniques such as geodesic distance calculation, dilation, occlusion detection and correction, and convex hull segmentation to efficiently track objects in depth images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If connected-component labeling is applied to three-dimensional depth images, then object identification and tracking capability is improved, but processing time and computational complexity increase significantly
Solution Approach 1:
The patent segments the depth image processing into distinct stages: initial labeling in screen dimensions, contour extraction, and selective depth dimension labeling. This segmentation allows the system to process only relevant portions of the data in full three-dimensional complexity, reducing overall computational burden while maintaining object identification accuracy.
Solution Approach 2:
The patent initially performs connected-component labeling in two dimensions (screen space) to identify potential objects, then selectively extends labeling to the third dimension (depth) only where necessary. This dimensional approach reduces computational complexity by avoiding full 3D processing of the entire image while still achieving reliable object tracking.
2Reliability
If connected-component labeling is applied to three-dimensional depth images, then object tracking capability is improved, but battery consumption increases excessively
Solution Approach 1:
The patent divides the processing workload into segments: initial 2D labeling that identifies candidate objects, contour extraction that defines object boundaries, and selective 3D labeling that tracks only identified objects. This segmentation reduces the total computational load and corresponding energy consumption while preserving tracking accuracy.
Solution Approach 2:
The patent applies full three-dimensional connected-component labeling only partially - specifically, only to regions containing identified objects and their contours. Rather than processing the entire depth image in 3D, the system performs selective labeling where needed, reducing energy consumption while maintaining sufficient tracking capability.
3Measurement precision
If full three-dimensional connected-component labeling is performed, then object identification accuracy is improved, but computational complexity becomes excessive
Solution Approach 1:
The patent segments the labeling process into hierarchical stages: first labeling in screen dimensions to identify candidate regions, then extracting contours, and finally performing depth dimension labeling only on relevant regions. This segmentation maintains identification precision by thorough processing of target areas while reducing overall computational complexity through selective application.
Solution Approach 2:
The patent performs preliminary connected-component labeling in screen dimensions before extending to depth dimensions. This preliminary action identifies candidate objects and their contours, allowing subsequent depth-based labeling to focus only on relevant regions, thereby maintaining precision while reducing the scope of complex 3D computations required.
Data Source
AI summary
In accordance with some embodiments, connected-component labeling is performed in both the screen dimensions (which may be referred to as the x and y dimensions) and a depth dimension to label objects in a depth image. Then the contour of labeled blobs may be used to identify an object in the depth image. Using contours may be advantageous in some embodiments because it reduces the amount of data that must be handled and the extent of computations, compared to conventional techniques which use bit map based operations.