Multi-source Crowd Counting via Fourier Analysis and MRF
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for counting individuals in dense crowds are inefficient due to their inability to handle high densities and reliance on temporal information, leading to inaccurate counting and estimation in still images.
Innovation Solution
A computer-implemented method that divides images into patches, uses Fourier analysis, head detection, and interest-point based counting to estimate individual counts, incorporating confidence and scale to improve accuracy, and applies a multi-scale Markov Random Field to enforce global consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If existing crowd counting methods are used, then counting can be performed on low-to-medium density crowds, but they fail to handle high density crowds of hundreds or thousands of individuals
Solution Approach 1:
The image is divided into multiple overlapping patches, and each patch is processed independently to estimate local crowd density. This segmentation allows the system to handle high-density crowds by breaking down the complex global counting problem into manageable local regions, where density estimation is more reliable even when overall crowd density is very high.
Solution Approach 2:
The patent transforms the counting problem from direct object detection into a density estimation problem in the frequency domain using Fourier transform. By analyzing the frequency characteristics of crowd patterns rather than individual objects, the system can accurately estimate counts in high-density scenarios where individual detection fails.
2Difficulty of detecting and measuring
If person detection methods are used for counting, then individual localization can be achieved, but detection becomes difficult due to severe occlusion and clutter in dense crowds
Solution Approach 1:
The patent extracts frequency domain characteristics from the image using Fourier transform, separating the crowd density information from the complex visual clutter and occlusions. By working in the frequency domain rather than the spatial domain, the system can estimate crowd counts without needing to detect or localize individual persons, thus avoiding the difficulties of occlusion and clutter.
Solution Approach 2:
The patent replaces the mechanical approach of individual person detection and localization with a field-based frequency analysis approach. Instead of tracking individual objects through occlusions, the system uses Fourier analysis to capture the overall spatial frequency patterns of the crowd, providing robust counting performance in dense scenarios.
3Measurement precision
If Fourier transform is used for texture analysis, then texton counts can be estimated, but the transform is not useful for localizing repeating elements in irregular arrangements
Solution Approach 1:
The patent applies Fourier analysis locally to each image patch rather than globally to the entire image. This local application allows the frequency analysis to capture crowd patterns in specific regions while being insensitive to irregular arrangements and occlusions. Each patch's frequency characteristics provide localized density estimates that are then combined to give the overall crowd count.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The method effectively estimates the number of individuals in extremely dense crowds, achieving improved accuracy and consistency across diverse crowd densities, outperforming existing techniques in terms of mean and normalized absolute difference errors.
Implementation Method 1
Fourier transform/analysis has been used extensively in texture analysis, and specifically in crowd analysis. Given geometrically arranged texture elements, the Fourier transform can provide reliable estimates of the texton counts.
Data Source
AI summary
A method for counting individuals in an image containing a dense, uniform or non-uniform crowd. The current invention leverages multiple sources of information to compute an estimate of the number of individuals present in a dense crowd visible in a single image. This approach relies on multiple sources, such as low confidence head detections, repetition of texture elements (using SIFT), and frequency-domain analysis to estimate counts, along with confidence associated with observing individuals in an image region. Additionally, a global consistency constraint can be employed on counts using Markov Random Field. This caters for disparity in counts in local neighborhoods and across scales. The methodology was tested on a new dataset of fifty (50) crowd images containing over 64,000 annotated humans, with the head counts ranging from 94 to 4,543. Efficient and accurate results were attained.