Pedestrian Recognition Model Training via Feature Intensity Occlusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional pedestrian recognition models in video surveillance have low accuracy due to limited training data and poor generalization performance, despite data augmentation methods like flipping and cropping, which do not significantly improve accuracy.
Innovation Solution
An image processing method that generates a feature intensity image indicating pixel importance, where a preset window is used to occlude regions in training images, creating new images to update the recognition model, improving model accuracy by emphasizing critical pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If conventional data augmentation methods (flipping, cropping, pixel perturbation) are used to increase training data quantity, then the quantity of training data is improved, but the model accuracy and generalization performance remain poor
Solution Approach 1:
The patent extracts and identifies critical/important pixels from training images using feature intensity images and gradient analysis. By separating important pixels from non-important pixels, the method focuses augmentation operations only on critical regions, thereby improving model accuracy while using the same quantity of training data more effectively
Solution Approach 2:
The patent applies different treatment to different parts of the image based on their importance. Important pixels (identified through gradient analysis and feature intensity) receive focused augmentation attention, while non-important pixels are handled differently. This local differentiation improves the quality of augmented data and subsequent model training
2Adaptability or versatility
If more training data is collected to improve model generalization, then the generalization performance should be improved, but the existing data augmentation methods fail to achieve significant accuracy improvement
Solution Approach 1:
The patent performs preliminary analysis to identify important pixels and regions before data augmentation. By pre-processing images to determine which pixels are critical for recognition, the method prepares the data in a way that subsequent augmentation operations can focus on improving generalization in the most relevant areas, rather than applying uniform augmentation across entire images
Data Source
AI summary
An image processing method, a related device, and a computer storage medium are provided. The method includes: obtaining a feature intensity image corresponding to a training image, where an intensity value of a pixel in the feature intensity image is used to indicate importance of the pixel for recognizing the training image, and resolution of the training image is the same as resolution of the feature intensity image; and occluding, based on the feature intensity image, a to-be-occluded region in the training image by using a preset window, to obtain a new image, where the to-be-occluded region includes a to-be-occluded pixel, and the new image is used to update an image recognition model. According to the embodiments of the present application, a prior-art problem that a model has low accuracy and relatively poor generalization performance because of limited training data can be resolved.