Object Search in Video Streams Using Feature Topology
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for searching and analyzing data in video streams, such as automatic face recognition and object tracking in CCTV systems, are limited by demanding image quality and shooting conditions, and lack the ability to search by object class or color characteristics, making them resource-intensive and restrictive in application.
Innovation Solution
A method that classifies objects in video frames, separates them from the background, creates a topology of their structure and features, and uses these features to efficiently search for specific objects across multiple cameras, allowing for precise and fast object location by setting search criteria based on object characteristics like color and structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face recognition and tracking methods are used in CCTV systems, then identification of persons can be performed, but demanding requirements are imposed on image quality and shooting conditions, significantly limiting the range of applications
Solution Approach 1:
The patent changes the parameters used for object identification from relying on specific facial features to using general object characteristics such as color, shape, and size. This allows the system to identify objects under varying image quality and shooting conditions, expanding application range while maintaining identification capability
Solution Approach 2:
The system is designed to identify multiple types of objects (not just human faces) using unified color and shape characteristics. This universal approach allows the same methodology to be applied across different object types and camera conditions, increasing versatility without sacrificing precision
2Productivity
If object search is performed using trajectory mapping and two-dimensional space analysis, then search speed can be accelerated, but the ability to search by color characteristics and object class is lost, and system complexity increases
Solution Approach 1:
The patent extracts and utilizes color characteristics and object class information directly from video frames without requiring complex trajectory mapping systems. By focusing on intrinsic object properties rather than spatial relationships between multiple cameras, the system maintains search speed while reducing complexity
Solution Approach 2:
Instead of requiring multiple physical cameras to capture trajectories, the system creates virtual representations of object characteristics (color, shape, size) that can be searched independently. This copying approach allows flexible searching by object attributes without the hardware complexity of multi-camera trajectory systems
3Reliability
If face identification and clothes analysis are used to track persons, then tracking can be performed even with similar appearances, but requirements on image quality are imposed and object characteristics depend on shooting angle
Solution Approach 1:
The patent changes from using detailed facial features and clothing patterns to using fundamental geometric parameters (shape, color, size) that remain consistent across different shooting angles. This parameter transformation maintains tracking reliability while eliminating angle-dependent limitations
Solution Approach 2:
The system uses homogeneous color and shape characteristics that are invariant to shooting angle, rather than heterogeneous detailed features like facial expressions or clothing textures. This homogeneity ensures reliable tracking across varying camera positions and angles
Data Source
AI summary
The claimed invention relates to the data processing industry, in particular, to the methods and systems of searching and analyzing data for finding an object in a video stream. The technical result is enhancing the precision of determining the sought object in a video stream, as well as increasing the speed of locating the object by means of performing a classification of objects and determining their features present on the video stream frames.