Two-Stage Object Detection for Reducing False Negatives
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated security systems face challenges in accurately detecting visual objects in videos due to limited computational resources on camera devices, leading to false negatives and missed classifications, which can be inefficient and require manual user input for correction.
Innovation Solution
A method and system that utilizes a first object detector on a camera device to identify potential missed object detections, which are then sent to a cloud server with more computational resources for further processing by a second object detector, allowing for user input to confirm and label interesting objects, thereby improving detection accuracy and reducing false negatives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a first object detector with limited computational resources is used on camera devices, then device complexity and cost are reduced, but detection accuracy deteriorates leading to false negatives
Solution Approach 1:
The detection system is segmented into two parts: a lightweight first object detector deployed on edge devices (cameras) for initial processing, and a more powerful second object detector deployed in the cloud for re-processing uncertain cases. This segmentation allows each component to have appropriate computational resources for its function, resolving the contradiction between device complexity and detection accuracy.
Solution Approach 2:
A cloud server acts as an intermediary between the first object detector and the final detection results. The cloud server receives video data from the first detector, applies a confidence threshold filter, and sends uncertain cases to the second object detector for re-processing. This intermediary mechanism allows the system to leverage both edge and cloud computational resources optimally.
2Reliability
If manual user input is required for correcting missed detections, then detection accuracy can be improved, but productivity and efficiency deteriorate
Solution Approach 1:
The system performs self-correction by automatically identifying false negatives through the two-stage detection process. The first object detector continuously monitors video streams, and when it encounters uncertain cases (low confidence scores), it automatically triggers re-processing by the second object detector in the cloud, eliminating the need for manual intervention while maintaining high accuracy.
Solution Approach 2:
The system implements feedback through the cloud server that receives detection results from the first object detector, evaluates their confidence scores, and provides corrective feedback by re-processing uncertain cases with the second object detector. This automated feedback loop continuously improves detection accuracy without requiring manual review.
3Reliability
If video data is sent to cloud server for re-processing, then detection accuracy is improved by leveraging additional computational resources, but loss of time and energy consumption increase
Solution Approach 1:
Instead of sending all video data to the cloud for re-processing (excessive action), the system applies partial action by selectively re-processing only those video segments where the first object detector achieved low confidence scores. This selective approach minimizes time loss and energy consumption while maintaining the benefit of improved detection accuracy for uncertain cases.
Data Source
AI summary
Reducing false negatives and finding new classes in object detectors is disclosed. Also disclosed is a method that includes grouping together a plurality of cropped image portions from region proposals based on image properties. The method also includes receiving first user input that either establishes that the grouped together, cropped image portions relate to uninteresting objects or establishes that the grouped together, cropped image portions relate to an object class of interesting objects. The method also includes obtaining, only when the grouped together, cropped image portions relate to the object class of interesting objects, second user input that includes an object label corresponding to the object class of interesting objects.