Face Detection via CNN Data Augmentation and End-to-End Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face detection algorithms face challenges with blurring, shielding, and excessively small faces, leading to inefficient detection processes in surveillance videos, which hinder real-time automation and manual labor liberation in security systems.
Innovation Solution
A fast side-face interference resistant face detection method utilizing convolutional neural networks and parameter sharing strategies, including data preprocessing through mirror symmetry and Gaussian filtering, and end-to-end training with region-based fully convolutional networks for improved detection speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional face detection algorithms are used, then detection can be performed on standard faces, but detection accuracy deteriorates when facing blurring, shielding, or excessively small faces
Solution Approach 1:
The patent applies preliminary action by performing data augmentation through mirror symmetry and Gaussian filtering operations on training images before model training. This preprocessing step creates enhanced training data that prepares the model to handle adverse conditions like blurring and shielding, improving detection accuracy without requiring changes to the detection algorithm itself during deployment.
Solution Approach 2:
The patent employs parameter changes by modifying image parameters during preprocessing - specifically applying Gaussian filtering to simulate and counteract blurring effects, and using mirror symmetry to create varied training samples. These parameter transformations in the training phase enable the model to maintain high detection accuracy under various adverse conditions.
2Reliability
If detection algorithms process all surveillance videos manually, then comprehensive search can be performed, but time consumption increases significantly
Solution Approach 1:
The patent replaces the mechanical manual search process with an automated convolutional neural network-based detection system. The end-to-end trained model automatically processes surveillance videos, extracting and classifying face features without human intervention, thereby eliminating manual search time while maintaining comprehensive detection coverage through automated batch processing.
Solution Approach 2:
The detection system implements self-service by autonomously processing surveillance videos without requiring manual operation. The trained model automatically performs face detection, extracts features, and classifies results, enabling the system to serve itself in completing comprehensive video searches without human time investment while maintaining reliability through consistent automated processing.
3Measurement precision
If complex detection models are used to improve accuracy, then detection precision increases, but processing speed decreases
Solution Approach 1:
The patent merges multiple functions into a single end-to-end trained convolutional neural network model. By combining feature extraction, candidate region generation, and classification into one integrated model, it eliminates the need for separate processing stages, thereby maintaining high detection accuracy while improving processing speed through unified computation and parameter sharing across all detection tasks.
Solution Approach 2:
The detection model achieves universality by being trained to handle multiple detection scenarios simultaneously - standard faces, blurred faces, shielded faces, and small faces - within a single model framework. This multi-functional capability allows the system to maintain high accuracy across diverse conditions without requiring multiple specialized models, thus preserving detection speed while achieving comprehensive precision.
Data Source
AI summary
Disclosed is a fast side-face interference resistant face detection method, in which a user selects an ordinary image, uses a deep neural network to extract image features, and then determines an exact location of a face. A training method for face detection uses a pure data-driven manner, uses an ordinary face image and a face boundary box as inputs, uses mirror symmetry and Gaussian filtering to perform data augmentation, and uses migration learning and hard example mining to enhance training effects. After a face image is read, the image is firstly scaled, and then placed into the deep neural network to extract features, and generate a plurality of face likelihood boxes and confidence scores of the face likelihood boxes, and finally the most appropriate face likelihood box is selected in a non-maximum suppression manner. No specific requirements are set on an angle of the face image, and a detection effect of a side face is still very obvious. In addition, the detection method above is simple, employs an end-to-end detection manner, and can be applied to a real-time environment.