Crowd Segmentation via Patch Affinity Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer vision systems for crowd segmentation in crowded areas face challenges in accurately distinguishing individuals from non-person objects and optimizing segmentations based on local image information, leading to inefficiencies in tracking individuals.
Innovation Solution
A method for globally optimizing crowd segmentation by initializing hypotheses based on input images, dividing them into patches, calculating affinity measures using whole body classifiers, and optimizing patch assignments through an EM algorithm, considering motion and color consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional computer vision systems are used for crowd segmentation, then the system can detect and track individuals in crowds, but the accuracy of distinguishing individuals from non-person objects is insufficient
Solution Approach 1:
The input image is divided into multiple patches, and each patch is independently evaluated against multiple hypotheses using affinity measures. This segmentation allows fine-grained analysis of local regions to distinguish individuals from non-person objects more accurately.
Solution Approach 2:
The system calculates affinity measures between patches and hypotheses based on local image characteristics such as motion consistency and color consistency. Each patch is evaluated using relevant local features rather than treating the entire image uniformly, improving the ability to distinguish individuals from non-person objects in specific local regions.
2Manufacturing precision
If global optimization is applied to crowd segmentation, then the segmentation can be optimized based on local information, but the computational complexity increases
Solution Approach 1:
By dividing the input image into patches and evaluating each patch against multiple hypotheses independently, the system reduces the complexity of global optimization. The affinity measures are calculated locally for each patch-hypothesis pair, making the overall optimization more manageable.
Solution Approach 2:
The system uses affinity measures that incorporate parameters such as motion consistency and color consistency to evaluate patch-hypothesis relationships. These parameters allow the system to optimize segmentation by adjusting weights and thresholds in the affinity calculation, balancing optimization precision with computational efficiency.
3Measurement precision
If multiple hypotheses are initialized and evaluated, then the tracking accuracy can be improved, but the processing time increases
Solution Approach 1:
Multiple hypotheses are initialized and affinity measures are pre-calculated for patch-hypothesis pairs before final assignment. This preliminary evaluation allows the system to efficiently assign patches to hypotheses without requiring extensive real-time computation, improving tracking accuracy while controlling processing time.
Solution Approach 2:
The system creates multiple hypothesis representations that can be evaluated against patches using affinity measures. By preparing and storing these hypothesis models in advance, the system can quickly assign patches to the most appropriate hypotheses without repeated complex computations during tracking.
Data Source
AI summary
Aspects of the disclosure provide a method for crowd segmentation that can globally optimize crowd segmentation of an input image based on local information of the input image. The method can include receiving an input image of a site, initializing a plurality of hypothesis based on the input image, dividing the input image into a plurality of patches, calculating an affinity measure of one or more patches to a hypothesis based on a partial response of the patches to a whole body classifier of the hypothesis that includes a combination of weak classifiers, and optimizing assignments of the plurality of patches to the plurality of hypothesis based on the affinity measures of the plurality of patches to the plurality of hypothesis.