Density Image Generation for Occupancy Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current behavioral analysis systems face challenges in effectively representing the attractiveness of objects of interest by accurately counting and tracking individuals in observation zones, especially in crowded environments with occlusions and varying camera angles, which complicates the generation of reliable density images.
Innovation Solution
A method for generating a density image of an observation zone over a given time interval involves extracting a fixed background, detecting 'blobs' and individuals using a three-dimensional model, and incrementing elementary surface intensity levels based on occupancy rates, with head detection utilizing a Canny filter, distance transform, and watershed algorithm, and optimizing individual identification through an appearance model with ellipses, addressing occlusions and camera distortions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional behavioral analysis algorithms are used for detecting and counting individuals, then facial detection and eye tracking can be performed, but accurate counting and tracking become difficult in crowded environments with occlusions and varying camera angles
Solution Approach 1:
The patent transforms the detection problem from 2D image space to 3D spatial representation. By creating a three-dimensional model of the observation area and mapping detected blobs to 3D positions, the system can distinguish overlapping individuals in depth, enabling accurate counting even when individuals are occluded in the 2D image plane.
Solution Approach 2:
The patent segments the detection task into distinct components: blob detection in image space, background subtraction to isolate moving objects, and 3D spatial segmentation to separate overlapping individuals. This multi-stage segmentation approach allows the system to handle crowded environments by processing detection at multiple levels of abstraction.
2Measurement precision
If multiple hypotheses of individual positions are generated within a three-dimensional model, then accurate individual detection can be achieved, but processing complexity increases
Solution Approach 1:
The patent performs preliminary background extraction to create a static background model before processing individual frames. This preliminary action separates the static environment from moving objects, simplifying subsequent blob detection and reducing the complexity of real-time processing by focusing computational resources only on dynamic elements.
Solution Approach 2:
The patent introduces a three-dimensional spatial model as an intermediary between 2D image detection and final individual positioning. This intermediary representation allows the system to generate and evaluate multiple position hypotheses in a structured 3D space, making the complex positioning task more manageable through systematic spatial reasoning.
3Reliability
If head detection is performed using contour analysis with Canny filter and watershed algorithm, then head positions can be detected without arbitrary threshold values, but processing time increases
Solution Approach 1:
The patent replaces traditional threshold-based detection methods with a watershed algorithm that uses gradient information and contour analysis. This substitution eliminates the need for arbitrary threshold selection, providing more reliable and consistent head detection across varying lighting and viewing conditions, though at increased computational cost.
Data Source
Figure 1~2
Figure 3~5
Figure 6~7
AI summary
The invention relates to a method for generating a density image of an observation zone over a given time interval, in which method a plurality of images of the observation zone is acquired, for each image acquired the following steps are carried out: a) detection of zones of pixels standing out from the fixed background of the image, b) detection of individuals, c) for each individual detected, determination of the elementary surface areas occupied by this individual, and d) incrementation of a level of intensity of the elementary surface areas thus determined in the density image.