Probability Distribution Construction for Subject Detection in Security Cameras
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Security cameras face challenges in recording moving images over long periods due to storage limitations, leading to unclear images and difficulties in identifying individuals, especially when compressing is applied, and there is a need for precise detection of persons facing or with their backs to the camera.
Innovation Solution
A method and apparatus that construct an appearance probability distribution of a subject's size based on multiple images from a fixed camera, allowing for high-precision detection by extracting similar subject shapes, calculating probability distributions, and integrating data from varying camera angles and focal distances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of stationary object
If a compressing process is applied to moving images for long-term recording, then storage capacity is extended, but image clarity deteriorates making precise identification difficult
Solution Approach 1:
The system performs preliminary detection of face portions and head portions in moving images before compression. By identifying these regions of interest in advance, the system can apply different compression strategies - preserving quality in detected regions while allowing higher compression in non-critical areas, thus maintaining identification precision over long recording periods
Solution Approach 2:
The patent implements local quality control by distinguishing between face portions, head portions, and other regions. Different compression rates are applied to different regions - face and head portions receive lower compression to maintain identification quality, while other areas undergo higher compression, resolving the contradiction between long-term storage and precise identification
2Measurement precision
If face detecting technique is used to detect face portion, then identification of persons facing camera is improved, but detection of persons with back directed to camera is not achieved
Solution Approach 1:
The system enhances the traditional face detection technique by adding head portion detection functionality. The detection apparatus can identify both face portions (for persons facing the camera) and head portions (for persons with backs directed to camera), making the system universal for detecting persons in various orientations without requiring multiple separate systems
Solution Approach 2:
The detection system segments the detection task into two distinct functions: face portion detection for identifying persons facing the camera, and head portion detection for tracking persons with backs to the camera. This segmentation allows each detection module to be optimized for its specific purpose while working together to provide comprehensive person detection capability
3Measurement precision
If detection precision is improved by considering environment information, then erroneous detection is suppressed, but system complexity increases
Solution Approach 1:
The system introduces an intermediary probability distribution calculation module that processes environment information. Instead of directly complex image processing, the system calculates probability distributions of subject sizes based on environmental context, which then guides the detection process. This intermediary approach improves detection precision while managing system complexity through modular design
Data Source
AI summary
A probability distribution constructing method extracts a subject shape similar to a subject of a specific type repeatedly appearing in various sizes in plural images obtained by repeatedly photographing a field using a fixedly disposed camera, from plurality images, in accordance with a size of the similar subject shape and positional information of the camera on a view angle. Subsequently, the probability distribution constructing method determines the similar subject shape, and calculates an appearance probability distribution of the size of the subject, and detects the subject using the appearance probability distribution.