Adversarial Image Masks for False Object Detection Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems in cameras suffer from false positive detections due to background objects, such as trees and mailboxes, which are incorrectly identified as objects of interest, leading to unnecessary alarms.
Innovation Solution
An adversarial mask is generated to target and modify pixels in images to prevent the detection of false alarm background objects (FABOs) while minimizing impact on true positive detections, using an iterative optimization process based on multiple images under varying conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional object detection models are used to detect objects in monitoring systems, then true positive detections are achieved, but false positive detections occur due to background objects like trees and mailboxes
Solution Approach 1:
The patent applies preliminary anti-action by generating adversarial masks that preemptively counteract the detection of false alarm background objects. The system identifies FABOs in training images and creates masks that modify pixel values to prevent these background objects from being detected as positive objects, thereby eliminating false positives before they occur in the monitoring system
Solution Approach 2:
The patent implements local quality by applying different treatments to different regions of the image. Adversarial masks are generated specifically for regions containing FABOs while leaving regions with potential true positive objects unchanged. This selective modification ensures that only background objects are affected, preserving the detection accuracy for genuine objects of interest
2Object-generated harmful factors
If pixel values are modified to reduce false positive detections, then false alarm background objects are prevented from being detected, but the detection of true positive objects may be affected
Solution Approach 1:
The patent introduces an intermediary element - the adversarial mask - that mediates between the input image and the object detection model. The mask selectively modifies only the FABO regions while preserving true positive object regions, acting as a filter that prevents false positives without interfering with genuine object detection. The mask serves as a bridge that allows the system to reduce false alarms while maintaining detection accuracy
3Adaptability or versatility
If adversarial masks are generated using multiple images under varying conditions, then robustness against different background objects is improved, but the processing complexity increases
Solution Approach 1:
The patent applies preliminary action by generating adversarial masks during an offline training phase using multiple images captured under varying conditions (different times, lighting, weather). This pre-processing step creates robust masks that adapt to diverse background objects before deployment. During actual monitoring operation, the pre-generated masks are applied directly without requiring complex real-time processing, thus achieving robustness while minimizing operational complexity
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving multiple images from a camera, each image of the multiple images representative of a detection of an object within the image. For each image of the multiple images the methods include: determining a set of detected objects within the image, each object defined by a respective bounding box, and determining, from the set of detected objects within the image and ground truth labels, a false detection of a first object. The methods further include determining that a target object threshold is met based on a number of false detections of the first object in the multiple images, generating, based on the number of false detections for the first object meeting the target object threshold, an adversarial mask for the first object, and providing, to the camera, the adversarial mask.