Video Conferencing Person Detection With Reflection False-Positive Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Video conferencing systems incorrectly identify reflections of people off glass walls or other reflective surfaces as real individuals, leading to undesirable outcomes such as miscounting people, misidentifying speakers, and incorrect zoom adjustments.
Innovation Solution
Implementing three filters: a confidence threshold filter, a similarity filter, and a delay filter to distinguish between real and reflected objects by assigning confidence scores, comparing sub-images for similarity, and delaying the use of newly detected objects to ensure they are not reflections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If detection software is used to identify sub-images of objects in video streams, then tracking and detection functionality is enabled, but false positive identifications occur where reflections are mistakenly detected as real people
Solution Approach 1:
The patent segments the detection process into multiple independent filtering stages: confidence score filtering, similarity comparison filtering, and temporal consistency filtering. Each filter processes detection results independently and sequentially, allowing the system to maintain automated detection while systematically eliminating false positives through layered validation.
Solution Approach 2:
The patent introduces confidence scores as an intermediary metric between raw detection and final object identification. This intermediary layer provides a quantitative measure of detection reliability, allowing the system to filter out low-confidence detections (likely reflections) before they are processed further, thereby improving overall detection accuracy.
2Reliability
If multiple filtering operations are applied to remove false detections, then detection accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The patent applies confidence score filtering as a preliminary operation before more computationally intensive similarity comparisons. By quickly eliminating low-confidence detections first, the system reduces the number of objects requiring detailed analysis, thereby minimizing overall processing time while maintaining high detection accuracy.
Solution Approach 2:
The patent implements a tiered filtering approach where not all detections undergo all filtering stages. Detections are processed through confidence filtering, and only those passing this initial filter proceed to similarity comparisons. This partial application of filtering operations reduces computational overhead while still achieving high accuracy for the most promising detections.
3Reliability
If confidence scores are used to filter sub-images, then false positives are reduced, but some true detections with lower confidence may be incorrectly discarded
Solution Approach 1:
The patent merges confidence score filtering with similarity comparison filtering in a coordinated manner. Detections that pass the confidence threshold are further validated through similarity comparisons with previously detected objects. This combination ensures that low-confidence detections are not discarded in isolation but are evaluated in context with temporal and spatial relationships, reducing false negatives.
Solution Approach 2:
The patent implements feedback through temporal consistency checking, where detection results from previous frames inform the evaluation of current frame detections. Detections are compared against historical data, and consistent patterns across multiple frames provide reinforcing evidence that validates even lower-confidence detections, preventing their premature discarding.
Data Source
AI summary
A method including detecting, in a digital image, a set of sub-images matching a selected object type. The method also includes generating a first confidence score that a first sub-image in set of sub-images matches a selected object type. The method also includes generating a second confidence score that a second sub-image in set of sub-images matches the selected object type. The method also includes generating a similarity measure by comparing the first sub-image to the second sub-image. The method also includes removing, responsive to the similarity measure exceeding a similarity threshold value and the first confidence score exceeding the second confidence score, the second sub-image from the set of sub-images. The method also includes processing, after removing, the digital image using the set of sub-images.