Safety helmet wearing detection method in complex scene
A technology of complex scenes and detection methods, applied in the field of deep learning and target detection, can solve the problems of small target factors, small size of individuals on the screen, mutual occlusion, etc., to improve the detection effect, ease the labeling work, and reduce the effect of information loss.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0073] A safety helmet wearing detection method in a complex scene, comprising the following steps:
[0074] Step A, according to the characteristics of the large size change of the detection target in the picture in complex scenes, design four feature scales for helmet detection;
[0075] On the basis of the YOLO v5 network detection of the neck and head, the fourth detection scale with a smaller receptive field is added to enhance the detection effect on small targets. The four detection scales are 13×13, 26×26, and 52 respectively. ×52,104×104, compared with the original three scales, it has a larger scale detection range;
[0076] During YOLO v5 training, the objective function of its bounding box regression and the real value B gt It is related to the predicted value B, and its calculation is shown in formula (1),
[0077]
[0078] where d is the center c of the true value gt and the distance between the center c of the predicted frame, l is the diagonal length of t...
Embodiment 2
[0130] This embodiment adopts the safety-helmet-wearing-dataset of network open source hard hat data set and data expansion picture, before targeted data expansion (DA), comprise altogether 9047 normal hats wearing hard hats and 9082 hats not wearing hard hats For the negative class person, the two categories are randomly divided into training set and test set according to the ratio of 8:2 to train and test the network. After targeted data expansion, the number of person classes increases to 35531, and the test set remains unchanged. . In order to verify the effectiveness of the changes proposed by the present invention, in this embodiment, the original YOLO v5 network is selected as the baseline, the fourth detection scale (FS) is added in turn, the attention mechanism (SB) is introduced, and the targeted data expansion ( DA), transfer learning (PT), tested on the same test set, and evaluated the model from two aspects of accuracy and speed. The experimental environment is sh...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


