Blob Representation Using Interior Points for Object Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer vision systems face difficulties in reliably tracking foreground objects that overlap or merge, as existing blob representation methods based on centroids often result in ambiguity and loss of object identity when objects interact, leading to inaccurate tracking.
Innovation Solution
Introducing a robust blob representation that allows multiple centers of mass, referred to as interior points, which are computed using distance images and non-maximum suppression, enabling more accurate tracking of objects that interact by minimizing the impact of overlap on relative location calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single centroid is used to represent a blob, then the representation is simple and computationally efficient, but tracking reliability deteriorates when objects overlap or merge
Solution Approach 1:
The blob representation is segmented from a single centroid to multiple interior points. Each interior point represents a distinct region within the blob, allowing the system to maintain separate object identities even when blobs merge. This segmentation enables the tracking system to distinguish between multiple objects that would otherwise be represented by a single centroid, thereby improving tracking reliability without excessive complexity increase.
Solution Approach 2:
The representation transitions from a single-point (0-dimensional) centroid to multiple points distributed within the blob (adding spatial dimensionality). This dimensional expansion allows the system to capture the spatial distribution of mass within the blob, providing more information for reliable tracking while maintaining computational efficiency through the use of distance transform algorithms.
2Reliability
If multiple interior points are used to represent a blob, then tracking reliability improves during object overlap, but computational complexity increases
Solution Approach 1:
The computationally intensive process of finding multiple local maxima through iterative optimization is replaced by an efficient distance transform algorithm. The distance transform computes the distance from each pixel to the nearest background pixel in parallel, and interior points are simply the local maxima of this distance map. This substitution dramatically reduces computational complexity while maintaining the ability to represent multiple interior points for reliable tracking.
3Use of energy by moving object
If a single centroid is used, then the system is computationally efficient, but measurement precision deteriorates when objects are in close proximity
Solution Approach 1:
The single centroid measurement is segmented into multiple interior point measurements. Each interior point provides an independent location measurement that is less susceptible to ambiguity when objects are close together. This segmentation of the measurement system improves location precision by distributing measurement points throughout the blob rather than relying on a single central point that may be ambiguous during overlap.
Solution Approach 2:
Interior points are pre-computed using distance transform before tracking operations. This preliminary computation stores the spatial distribution information that can be quickly referenced during tracking, avoiding the need for complex real-time calculations when objects are in close proximity. The pre-computed interior points provide ready-to-use precise location data that improves measurement precision without increasing real-time computational energy consumption.
Data Source
AI summary
A method of processing a video sequence is provided that includes receiving a frame of the video sequence, identifying a plurality of blobs in the frame, computing at least one interior point of each blob of the plurality of blobs, and using the interior points in further processing of the video sequence. The interior points may be used, for example, in object tracking.