SSD face detection method based on deep learning
A technology of face detection and deep learning, applied in the field of face detection, can solve the problems of inability to achieve real-time detection and slow detection speed, and achieve the effect of enhancing generalization ability and robustness, increasing diversity, and balancing category distribution
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Example Embodiment
[0043]Example 1
[0044]ReferenceFigure 1 ~ Figure 7 , Is the first embodiment of the present invention, and this embodiment provides a deep learning-based SSD face detection method, including:
[0045]S1: Perform candidate frame extraction, target detection and bounding box regression tasks based on the deep convolutional neural network for the face data to be tested. It should be noted that the deep convolutional neural network includes a convolutional layer, a pooling layer and a fully connected layer.
[0046]S2: Use the SSD strategy to discretize the output space of the target Bounding Box regression task into a priori box, and set a priori box of various aspect ratios and sizes corresponding to each position in each detection layer. What needs to be explained in this step is that the SSD strategy adopts the SSD network, which includes:
[0047]Use VGG16 as BackBone (pillar), and replace the sixth fully connected layer and seventh fully connected layer of VGG16 with the convolutional layer...
Example Embodiment
[0083]Example 2
[0084]The SSD network can detect multi-scale faces. The so-called multi-scale detection is to use different feature maps for face detection, and in each detected feature map, a priori box is set according to certain rules for multi-scale face matching. Generally speaking, The front feature map of the convolutional neural network is larger, and the receptive field corresponding to the feature unit is smaller. The convolutional layer at the back reduces the feature map through convolution and pooling to reduce the feature map, but it has a larger receptive field; therefore, The SSD algorithm sets default boxes of different scales to match the effective receptive fields of feature units of different feature maps, sets a small-scale prior box in the larger feature map in the front for small face detection, and sets a large in the smaller feature map in the back. Large face detection is performed on the priori box of the scale.
[0085]In layman's terms, the SSD300 algorithm ...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap