Multi-Camera Person Re-Identification for Duplicate-Free Counts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multi-camera video conferencing systems face challenges in accurately identifying unique participants due to overlapping camera fields of view, leading to duplicate detections and an inaccurate count of participants in the room.
Innovation Solution
A method involving machine learning models to detect human heads and generate unique signatures or embeddings for each participant, using techniques like Omni-Scale Network (OSNet) to associate images from different perspectives and eliminate duplicates, ensuring accurate participant counting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multiple cameras are used to provide coverage of all participants, then the coverage area is improved, but the accuracy of participant identification deteriorates due to duplicate detections
Solution Approach 1:
The patent segments the participant identification task by assigning unique IDs to each participant and tracking them across different camera views. Each camera detects participants independently, but the system divides and conquers the deduplication problem by using feature extraction and matching to identify which detections correspond to the same person across different cameras.
Solution Approach 2:
The patent introduces an intermediary identification system that mediates between multiple camera detections. This intermediary layer extracts features from each camera's detections, compares them using embedding similarity, and resolves duplicates by determining whether detections from different cameras refer to the same participant, thus maintaining identification accuracy across multi-camera coverage.
2Ease of operation
If single-camera systems automatically select optimal views, then the ease of operation is improved, but the capability to handle multi-camera overlapping views deteriorates
Solution Approach 1:
The patent makes the video conferencing system universal by designing it to handle both single-camera and multi-camera configurations. The same framework that automatically selects optimal views in single-camera systems is extended to multi-camera systems, where it additionally performs participant identification, tracking, and deduplication across multiple overlapping views, thus maintaining ease of operation while gaining multi-camera versatility.
3Measurement precision
If feature embedding techniques are used to identify unique participants, then the measurement precision is improved, but the computational complexity increases
Solution Approach 1:
The patent replaces complex computational participant identification mechanisms with a more efficient approach using pre-trained deep learning feature extractors. Instead of implementing custom complex algorithms for tracking and matching participants across cameras, the system substitutes these with established feature embedding techniques (like OSNet) that leverage transfer learning, thereby reducing the computational burden while maintaining high identification accuracy.
Data Source
AI summary
A multi-camera video conference call system is provided with a plurality of cameras connected together over a communication network to generate a corresponding plurality of input frame images taken from different perspectives of a video conference room, where the multi-camera video conference call system detects one or more human heads for any meeting participants captured in the input frame images, generates a head bounding box which surrounds each detected human head, extracts a body bounding box which surrounds the detected human head and at least an upper body portion of a meeting participant belonging to the detected human head, generates a participant identification feature embedding from each body bounding box, and performs person re-identification processing on all generated participant identification feature embeddings to determine a count of the meeting participants in the video conference room.