Scene-Customized Adversarial Masks for Camera False-Positive Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection models in cameras often produce false positive detections due to background objects, leading to unnecessary alarms and resource wastage.
Innovation Solution
An adversarial mask is generated to target and modify pixels associated with false alarm background objects (FABOs) to reduce false positive detections while minimizing impact on true positive detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection models detect all objects within field of view, then detection coverage is improved, but false positive detections increase
Solution Approach 1:
The patent segments the image processing by generating separate adversarial masks for different false alarm background objects (FABOs). Each FABO receives targeted pixel value adjustments through individual masks, allowing the system to address false positives locally without affecting the detection of other objects. This segmentation enables precise control over which regions are modified to reduce false alarms.
Solution Approach 2:
The adversarial mask technique applies local quality changes by modifying pixel values only in specific regions corresponding to FABOs. The mask adjusts brightness, contrast, or other pixel properties locally around detected false positive objects, thereby reducing their detectability as threats while leaving the rest of the image unchanged for accurate detection.
2Object-generated harmful factors
If adversarial mask modifies pixel values to reduce false positives, then false detection removal is improved, but impact on true positive detections may increase
Solution Approach 1:
The system dynamically adjusts adversarial mask parameters based on the specific characteristics of each FABO and its context in the image. The mask strength, pixel value adjustments, and region of influence are adapted in real-time to minimize false positives while preserving true positive detections. This dynamic approach allows the system to respond to varying detection scenarios optimally.
Solution Approach 2:
The patent implements a feedback mechanism where the object detection model's outputs are continuously monitored, and adversarial masks are generated or adjusted based on detected false positives. The system learns from detection results and refines mask parameters iteratively, using feedback from false alarm patterns to improve future mask generation and reduce both false positives and false negatives.
3Object-generated harmful factors
If multiple images are processed to optimize target regions, then false positive removal effectiveness is improved, but processing time increases
Solution Approach 1:
The system performs preliminary action by pre-processing multiple images to identify and optimize target regions before generating final adversarial masks. By analyzing patterns across multiple images in advance and pre-determining which regions are likely to produce false positives, the system reduces the computational burden during real-time detection and speeds up overall processing while maintaining effectiveness.
Solution Approach 2:
The patent merges information from multiple images by combining target region data across the image set to create composite adversarial masks. Instead of processing each image independently, the system integrates detection patterns from multiple images to identify consistent FABO locations and characteristics, thereby reducing false positives across all images while sharing computational work.
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.