Stereo People Counting via Disparity Gradient Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing stereoscopic people counting devices struggle to accurately distinguish individuals from other objects with similar heights and shapes, and are not adaptable to varying installation heights or lighting conditions, particularly in areas with variable lighting such as skylights, dimly lit spaces, or outdoors.
Innovation Solution
A method involving the generation of rectified images from two cameras, calculation of disparity images, and detection of people based on gradient analysis, using techniques like CENSUS and Hamming distance to identify corresponding pixels, with filtering and thresholding to enhance accuracy and adaptability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If distance threshold comparison is used to detect people, then the system is simple to implement, but it cannot distinguish individuals from other objects with similar heights and shapes
Solution Approach 1:
The patent segments the detection process into multiple stages: generating rectified images from stereo cameras, computing disparity maps, calculating gradients, and applying multi-criteria filtering. This segmentation allows complex analysis while maintaining systematic organization, resolving the contradiction between detection accuracy and implementation simplicity.
Solution Approach 2:
The patent transitions from 2D image analysis to 3D disparity-based analysis by computing depth information from stereo image pairs. This dimensional change enables distinction between objects at different depths, allowing accurate people detection even when they have similar heights and shapes in the 2D image plane.
2Adaptability or versatility
If fixed detection parameters are used, then the system is easy to install, but it is not adaptable to varying installation heights and lighting conditions
Solution Approach 1:
The patent implements dynamic adaptation by adjusting detection parameters based on installation height. The system calculates optimal parameters (such as disparity thresholds and gradient criteria) according to the actual mounting position, enabling the same device to adapt to different installation scenarios without requiring physical reconfiguration.
Solution Approach 2:
The patent changes detection parameters dynamically based on lighting conditions and installation height. By adjusting thresholds and processing parameters according to environmental conditions, the system maintains high detection accuracy across variable conditions without requiring complex hardware modifications.
3Measurement precision
If gradient analysis with multiple filtering steps is applied, then people detection accuracy improves, but processing time increases
Solution Approach 1:
The patent applies preliminary rectification and environmental image generation before disparity calculation. By pre-processing images to correct geometric distortions and enhance relevant features beforehand, the system reduces the computational burden during the main detection phase, balancing accuracy with processing efficiency.
Solution Approach 2:
The patent extracts and focuses on the most discriminative features (gradients in disparity images) while discarding redundant information. By concentrating computational resources on key features that directly indicate people presence, the system achieves high accuracy without excessive processing time.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates mainly to a method for counting people situated in a scene in a field of view of a stereoscopic appliance (1), characterised in that it comprises the following steps: generating a left image and a right image from two cameras (3) associated with optical systems (5); producing images, called left and right rectified images, wherein a point of the scene corresponds to pixels situated on a same line of the left rectified image and the right rectified image; identifying, for each pixel of an image, the corresponding pixel of the scene in the other image, and deducing an image therefrom, called an image of disparities wherein the value of each pixel corresponds to the distance, in a same line, between the row of corresponding pixels in the two images; calculating a gradient between each pixel of the image of disparities and a preceding pixel on a same line; and detecting a person according to the calculated gradients and counting the people detected in this way.