Face-gesture cooperative verification method based on deep learning detection

Abstract

The invention relates to a face-gesture cooperative verification method based on deep learning detection so that face and gesture detection can be completed by using an algorithm for a low-power-consumption device and thus an effect of face-gesture cooperative verification can be realized. Overall head-shoulder detection is carried out on an image reduced substantially by using a deep residual convolutional neural network; a face is detected based on a range with head-shoulder occurrence; and then a gesture is detected on the head and shoulder positions and then a trigger condition is verifiedbased on the detected face and gesture. Compared with the traditional gesture-based verification system, the face-gesture cooperative verification method has the following advantages: the stability is high; false triggering by a user and the out-of-order probability of the system are reduced substantially; the stability is improved; and the method can be applied to startup-shutdown verification of a mobile phone of a home appliance.

Description

technical field [0001] The present invention relates to technical fields such as artificial neural network and computer vision, and in particular to a method for verifying human face gesture cooperation based on deep learning detection. Background technique [0002] Static gesture recognition has important applications in visual communication, human-computer interaction, augmented reality and other fields. However, in practical applications, due to factors such as illumination and individual changes, it is difficult for simple gesture recognition to meet specific application requirements. In recent years, in the home appliance industry such as televisions, air conditioners, and air purifiers, the traditional method of relying on remote controls has become more and more clumsy, while the control method of relying on face plus gesture verification has become more and more popular in the high-end home appliance industry and the smart phone industry. Pay attention to. The face...

AI Technical Summary

Problems solved by technology

The current mainstream target detection algorithms such as SSD, R-FCN, mask R-CNN, etc. use popular deep network structures such as VGG-16, Resnet101, etc., and the time required to run it once using an ordinary notebook CPU cannot meet the real-time requirements. Devices with no GPU power consumption will cause memory overflow problems, which are far from meeting the requirements of low-power devices in the home appliance and mobile phone industries.

Furthermore, the low-power chips of these devices often run other programs at the same time in actual use, and have strict requirements on heat generation. Therefore, it is difficult to implement and implement common deep learning algorithms on these devices.

Images