Hand key point detection method based on deep learning

Abstract

The invention discloses a hand key point detection method based on deep learning. The method comprises a CPMs convolutional neural network and a GoogLeNet neural network model. In the CPMs convolutional neural network, the data layer in the step 1 and the last seven convolutional layers are reserved, and other layers in the step 1 are removed. In the GoogLeNet neural network model, front 13 network layers are selected, and a data layer and a max pool layer behind the inception (3b) are delected. The CPMs convolutional neural network and the GoogLeNet neural network model are combined and thebody attribute and the name of the top attribute of each layer are modified, so that the layers can be linked. The network model structure is improved, the accuracy of key point detection is improved,the cost (training time and the like) of model training is reduced, and the detection speed of model parameters is increased.

Description

technical field [0001] The invention relates to the field of hand key point detection, in particular to a deep learning-based hand key point detection method. Background technique [0002] Human skeleton key point detection is one of the basic algorithms of computer vision, and has played a fundamental role in research in other related fields, such as behavior recognition, person tracking, gait recognition and other related fields; specific applications are mainly concentrated in intelligent video Surveillance, patient monitoring system, human-computer interaction, virtual reality, human animation, smart home, athlete-assisted training, etc. Among them, hand key point detection is an important extension of human skeleton key point detection, and it is also a specific part of human skeleton key point detection. field of application. [0003] The Flowing Convnet in 2015 regards the detection of human skeleton key points as a detection problem, and the output is a heatmap. Thi...

AI Technical Summary

Problems solved by technology

[0003] Flowing Convnet in 2015 regards the detection of key points of human bones as a detection problem, and the output result is a heatmap. This model can only detect the key points of human bones in the upper body of the human body, and the detection range is limited.

However, the common shortcomings of these models are that there are many model parameters, the model training time is relatively long, and the accuracy of model detection still has room for improvement.

Images