Seatbelt Recognition Method Based on Convolutional Attention Mechanism
- 1 August 2019
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 187-192
Abstract
In order to solve the problem of low accuracy of the driver's seat-belt identification and neglecting co-pilot, a method of driver's seat-belt identification based on deep convolutional channel modeling is improved. Firstly, the characteristic parameters of redundant channels are reduced by means of deep convolutional feature channel compression. Then, the feature weight of the deep convolutional channel is re-calibrated to improve the attention weight of the feature of the safety belt and suppress the expression of the interference feature. Finally, the central loss function is used to assist feature training of the model, so as to reduce the difference of feature distribution within the class and make the learned features easier to be classified and recognized. The experimental results show that, compared with other algorithms, the improved deep convolution driver seat belt identification model in this paper improves the recognition accuracy, recall rate, and F value by 6.9%, 4.32 and 5.5% respectively, and greatly improves the accuracy and practicability.Keywords
This publication has 9 references indexed in Scilit:
- Lightweight Deep Neural Network Approach for Parking Violation Detection2018
- Squeeze-and-Excitation Networks2018
- Accurate seat belt detection in road surveillance images based on CNN and SVMNeurocomputing, 2018
- Deep Learning System for Automatic License Plate Detection and Recognition2017
- Image Classification by Selective Regularized Subspace LearningIEEE Transactions on Multimedia, 2015
- Real-Time Hand Gesture Detection and Recognition Using Bag-of-Features and Support Vector Machine TechniquesIEEE Transactions on Instrumentation and Measurement, 2011
- Image-based seat belt detection2011
- Multiple Vehicles Detection and Tracking based on Scale-Invariant Feature Transform2007
- Histograms of Oriented Gradients for Human Detection2005