Human body behavior recognition algorithm based on deep learning

Abstract

The invention provides a human body behavior recognition algorithm based on deep learning. The human body behavior recognition algorithm comprises: (1) preprocessing an input video segment; (2) constructing a network model RD3D; (3) defining a loss function, accuracy and operation of an optimizer; (4) training the network model comprises the following sub-steps: (41) initializing parameters; (42)enabling the learning rate to be 0.0001, and the bit size to be 16; (43) calculating the loss according to the forward propagation value of the RD3D model and the real label, and updating the weight parameter of the loss through back propagation; (44) training for 100 epoch, and then ending the training; (5) testing results. The method pursues the precision of the recognition algorithm from the perspective of features, solves a problem that a conventional algorithm seriously depends on a data set, reduces the sensitivity to the type of the data set, and can be applied to any behavior recognition data set.

Description

technical field [0001] The present invention relates to the technical field of computer vision, and more specifically, to a human behavior recognition algorithm based on deep learning. Background technique [0002] In recent years, with the rise of deep learning and other related technologies, deep neural networks have made breakthroughs in various fields such as computer vision. Due to its end-to-end training characteristics, deep learning can learn its common features from training data and fit a network suitable for the current task. At the same time, it is very easy to obtain massive data in today's society, which also provides convenient conditions for the application of deep learning to video understanding, recognition and other fields. [0003] In traditional methods, local features (such as HOG, HOF, and MBH) are mainly extracted, which requires strong prior knowledge. Although the appearance and motion information are considered, this information is only limited t...

AI Technical Summary

Problems solved by technology

Inputting the RGB image and the optical flow image into the network for learning also cannot be performed end-to-end, and the entire network has only seven layers, of which only three layers are used for feature extraction. Although the amount of calculation is small, the accuracy is very low;

Images