Digital Image Anomaly Detection With Adaptive Probability Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital image processing methods for anomaly detection in autonomous driving suffer from performance degradation due to environmental changes, such as lighting and weather conditions, leading to reduced accuracy in classifying or semantically segmenting digital images.
Innovation Solution
A computer-implemented method and device that utilize a trained model to process digital images by determining a subset of classes based on probability assignments, where the size of the subset indicates an anomaly score, and adaptively adjust thresholds to account for domain shifts in environmental conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a model is trained on digital images from multiple environmental conditions, then the anomaly detection performance is improved across varying environments, but the training data requirements and model complexity increase
Solution Approach 1:
The patent segments the environmental conditions into distinct domains (e.g., daytime, nighttime, rain, sunshine) and trains the model to handle each domain separately through domain adaptation techniques. This allows the model to specialize in each environmental condition while maintaining overall reliability across all conditions.
Solution Approach 2:
The patent changes the model parameters adaptively based on the detected environmental domain. By identifying the current environmental condition and adjusting model parameters accordingly, the system maintains high anomaly detection performance without requiring a completely different model for each condition.
2Measurement precision
If the model uses a fixed threshold for anomaly detection, then the detection process is simple and fast, but the accuracy degrades when environmental conditions shift from training data
Solution Approach 1:
The patent implements dynamic threshold adjustment based on the detected environmental domain. Instead of using a fixed threshold, the system adapts the threshold according to the current environmental conditions, maintaining high detection accuracy while adding only minimal complexity through domain classification and parameter selection.
3Productivity
If the model is trained only on specific environmental conditions, then the training process is simpler and faster, but the performance degrades when processing images from different environmental conditions
Solution Approach 1:
The patent creates a universal model that can handle multiple environmental conditions by training on diverse data from various domains. The model incorporates domain adaptation mechanisms that allow it to generalize across different environmental conditions, achieving both training efficiency and environmental versatility.
Solution Approach 2:
The patent introduces domain adaptation techniques as an intermediary layer between the model and different environmental conditions. This intermediary allows the model trained on specific conditions to effectively process images from other conditions by adjusting for domain shifts, maintaining both training simplicity and environmental adaptability.
Data Source
Figure 1
Figure 2
AI summary
Device and computer implemented method for processing a digital image for anomaly detection, wherein the method comprises providing (202) a model that is trained with a data set that comprises digital images to output probabilities that are assigned to classes in a set of classes for classifying or semantically segmenting the digital image depending on the digital image, determining (206) the probabilities for the digital image with the model, determining (208) a size of a sub-set of the set of classes depending on a sum of the probabilities that are assigned to the classes in the sub-set and depending on a first threshold, and detecting (210) an anomaly depending on the size, wherein the sum of the probabilities that are assigned to the classes in the sub-set is smaller than the first threshold, wherein determining (208) the size of the sub-set comprises determining the first threshold depending on a weighted sum of scores, wherein a score is weighted in the weighted sum depending on a likelihood ratio between a distribution of the probabilities for the digital image and a distribution of the probabilities of the digital images in the data set.