3D Head Model Matching for Person Detection in CCTV
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CCTV systems require multiple operators to monitor multiple cameras, leading to high error rates in detecting important events, and existing technologies like infrared sensors and single cameras struggle with accurate counting and discrimination of individuals, especially in dynamic environments with lighting changes.
Innovation Solution
A method and device using a three-dimensional camera to detect objects by generating depth information, identifying persons through 3D head model matching, and calculating features for situation detection, including crowd density and remaining person detection without relying on color information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multiple operators monitor multiple CCTVs, then more areas can be covered, but detection accuracy decreases due to human error
Solution Approach 1:
The patent replaces the mechanical human monitoring system with an automated intelligent detection system that uses image processing algorithms to automatically analyze CCTV footage, detect objects, and identify abnormal events, thereby eliminating human error while maintaining wide coverage
Solution Approach 2:
The system enables self-service detection by automatically processing video streams, identifying objects and events, and generating alerts without requiring human operators to continuously watch multiple screens, allowing the system to monitor itself and report abnormalities autonomously
2Productivity
If infrared sensors are used for counting, then rapid counting is possible, but accuracy deteriorates due to light disturbance and inability to discriminate objects
Solution Approach 1:
The patent merges the advantages of infrared sensing (rapid detection capability) with visual image processing (discrimination and identification capabilities) to create a hybrid system that achieves both high-speed counting and accurate object discrimination by combining data from multiple sensing modalities
Solution Approach 2:
The system changes the detection parameters by using visible light image processing instead of infrared detection, allowing for accurate discrimination of objects based on visual features while maintaining rapid processing speeds through optimized image analysis algorithms
3Device complexity
If single camera with color information is used, then simple implementation is achieved, but detection accuracy worsens due to sensitivity to lighting changes
Solution Approach 1:
The patent transitions from two-dimensional color image processing to three-dimensional depth information processing by utilizing depth maps or stereo vision, adding a spatial dimension that provides robust object detection independent of lighting conditions and color variations
4Measurement precision
If depth information is used for object detection, then overlapping objects can be separated, but device complexity increases
Solution Approach 1:
The patent applies depth information from a third dimension to separate overlapping objects in two-dimensional images, using depth maps or stereo disparity to distinguish objects that appear merged in the visual plane, thereby achieving accurate separation without requiring complex multi-camera setups
Data Source
AI summary
Provided is a detection method of a detection device. The detection device detects at least one object from an image by using depth information about the image obtained by a camera. The detection device identifies whether said at least one object is a person through three-dimensional (3D) head model matching, which matches a head candidate area of said at least one object with a 3D head model. The detection device calculates a feature for detection of a situation by using said at least one object when it is identified that said at least one object is the person.