Human action recognition is one of the core tasks in the field of computer aided driving. Considering that the auxiliary driving system requires high real-time, and the hardware requirements can not be too high, it is proposed to identify human behavior in a single image. Considering that the night illumination is insufficient, and the infrared camera receives the infrared radiation of the object, it can work at night without the influence of visible light. Therefore, we focus on the human behavior recognition in infrared images. According to the scale of the problem, we first use AlexNet with moderate network depth as the backbone network, then improve the network, modify the classification output layer of the network according to the classification number. After preprocessing the dataset to adapt to improve AlexNet, we trained and tested the network. The experimental results quantify the classification performance of the network. Experimental results show that the proposed algorithm mean average precision, average recall and F1 score are better than traditional methods.
Aiming at the problem of pedestrian behavior recognition in infrared images, a method based on Improved GoogLeNet is proposed. Firstly, by analyzing the application scenarios and the characteristics of common network models, GoogLeNet with better comprehensive performance is selected as the backbone network. Inspired by NIN, a kind of 1*1 convolution kernel structure is introduced to reduce the number of channels and significantly reduce the number of parameters. Then channel padding and resize to adapt to the network requirements for the training set and test set of the infrared image human behavior data set. Next, the fully connected layer and the classification output layer of the network are modified according to the number of behavior types contained in the data set. The convolution kernel and inception parameter in the pre-training network are introduced to accelerate the network training and improve the generalization ability of the network. Finally, the quantitative index is used to analyze the experimental results and judge the recognition performance of the network. Experimental results shows that the Mean Average Precision, Average Recall and F1 score obtained by the proposed algorithm are better than the traditional methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.