Data center-oriented human behavior attribute real-time detection method and system

Abstract

The invention provides a data center-oriented human body behavior attribute real-time detection method and system. The method comprises the following steps: acquiring attribute identification data sets of different human bodies in different environments of a data center, labelling the human body target frames and attribute labels in the attribute recognition data set, and dividing the labeled data set; preprocessing the divided data set pictures by adopting a splicing and cutting mode, then combining a human body target frame with attribute recognition to construct a human body attribute recognition integrated detection network, and performing transfer learning training on the detection network by improving a human body attribute loss function; and deploying the trained model to realize online reasoning and human body attribute prediction. Based on the method, the invention further provides a data center-oriented human behavior attribute real-time detection system. According to the invention, the loss function of the network is improved to eliminate the problem of imbalance of positive and negative samples in attributes, and guarantee is provided for high-precision intelligent identification.

Description

technical field [0001] The invention belongs to the technical field of image recognition in data centers and artificial intelligence, and in particular relates to a data center-oriented real-time detection method and system for human behavior attributes. Background technique [0002] With the continuous development of information technology and the continuous advancement of artificial intelligence technology, the intelligent development of data centers has become a new trend. Modular Data Center (MDC) is a new design concept adopted to adapt to the development trend of cloud computing, virtualization and centralized servers. It adopts a modular design concept, which can be quickly deployed and easily expanded. Operation and maintenance management is an important part of the data center. Monitoring the operation behavior of personnel in the data center can detect potential safety hazards in time, and provide early warning mechanisms in time to ensure the safety of operators a...

AI Technical Summary

Problems solved by technology

This method is relatively cumbersome in manual feature extraction, and the effect of human attribute recognition depends on the quality of the extracted features. In practical applications, it is affected by factors such as human appearance, behavior, posture, and illumination. The traditional machine learning method Difficult to apply

The main strategy of the human body attribute recognition method based on deep learning is to firstly identify the human body through the target detection method, and then identify the human body behavior attributes through an independent human body attribute recognition network or an associated human body attribute recognition network. The method of cascading is adopted, that is, human body recognition and behavior attribute recognition are regarded as two independent parts. This method has the following disadvantages: 1) The network method of multi-level series needs to be trained separately for the two-level network. And there is inconvenience in tuning; 2) Due to the existence of the cascade method at the inference end, it takes two parts of the network inference time, which is not conducive to the real-time requirements in the deployment of the embedded end; 3) The network identification results of the behavior attribute identification end are affected by Influence of Human Detection Network Recognition Results

Therefore, the traditional methods of manual on-site inspection or manual video surveillance are inefficient and prone to missing inspections; traditional machine learning and existing deep learning methods are difficult due to the influence of the external environment or network structure. Applied in the data center field

Images