Scene Geometry Calculation for Crowded Video Human Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance systems face challenges in automatically detecting humans in crowded scenes and settings with PTZ cameras, as existing methods rely on background subtraction or visible orthogonal directions, which fail when people occlude each other or when PTZ cameras zoom, making real-time human detection difficult.
Innovation Solution
A system that automatically calculates scene geometry by determining the vertical vanishing point and horizon based on dominant edge orientations and optical flow, allowing for real-time human detection without manual calibration, even in crowded scenes, by dividing video frames into blocks, detecting edges, and estimating scale and horizon.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If background subtraction method is used for human detection, then it works well for sparse scenes, but it fails in crowded scenes where people occlude each other
Solution Approach 1:
The patent segments the video frame into multiple blocks and processes each block independently to detect edges and determine dominant orientations. This segmentation allows the system to handle crowded scenes by processing local regions separately, avoiding the occlusion problems of global background subtraction methods.
Solution Approach 2:
The patent introduces an intermediary approach by detecting edges and calculating dominant orientations as intermediate steps to determine vanishing points and scene geometry. This intermediary process enables human detection without relying on background subtraction, making it effective in crowded scenes where objects occlude each other.
2Measurement precision
If manual calibration is performed after every camera movement, then scene geometry accuracy is maintained, but it becomes unrealistic and time-consuming
Solution Approach 1:
The patent implements self-service calibration where the system automatically determines scene geometry using edge detection and dominant orientation analysis. The system serves itself by autonomously calculating vanishing points and updating scene parameters without external manual intervention, even after camera movements or zoom operations.
Solution Approach 2:
The patent performs preliminary actions by continuously detecting edges and calculating dominant orientations in real-time video frames. This preliminary processing enables the system to maintain scene geometry accuracy automatically, preparing the necessary geometric information before human detection is performed.
3Measurement precision
If three orthogonal dominant directions method is used, then it can calculate human height, but it fails when orthogonal directions are not visible or camera zooms occur
Solution Approach 1:
The patent employs a dynamic approach where the system continuously adapts to camera movements by re-detecting edges and recalculating dominant orientations in real-time. This dynamic processing enables the system to maintain accurate scene geometry even when PTZ cameras pan, tilt, or zoom, unlike static methods that require visible orthogonal directions.
Solution Approach 2:
The patent changes parameters by calculating dominant orientations and vanishing points dynamically based on current video frame content. This parameter change approach allows the system to adapt to different camera positions and zoom levels, maintaining human detection accuracy without requiring visible three-dimensional orthogonal directions.
4Ease of manufacture
If background model is used for calibration, then it works for simple scenes, but it cannot isolate individual objects in crowded scenes
Solution Approach 1:
The patent segments the scene into multiple blocks and processes edges within each block independently. This segmentation enables the system to detect and isolate individual objects even in crowded scenes, as each block can be processed separately to identify local edge patterns and dominant orientations without being overwhelmed by the entire crowded scene.
Data Source
AI summary
A system or method for automatic calculation of scene geometry may include receiving, by a processor, a video stream including a plurality of video frames. A video frame may be divided into blocks, and a vertical vanishing point of the video stream may be determined based on a dominant orientation of detected edges in the blocks. A horizon and scale of the video stream may be estimated based on optical flow of pixels between two consecutive video frames. A set of human heads may be detected in a video frame of the video stream based on the estimated horizon and scale of the video stream.