Spatial gesture control method based on deep learning model cascading

Abstract

The invention provides a spatial gesture control method based on deep learning model cascading. The method comprises the steps of 1, collecting image data by a left camera and a right camera respectively; 2, performing palm detection on the acquired image data according to the deep learning model; 3, performing key point detection on a palm detection result; 4, carrying out refined secondary processing on a result after key point detection; 5, performing space coordinate calculation according to the palm detection result and the refined secondary processing result; and 6, performing space gesture recognition according to the space coordinates. Gesture interaction can be applied to the ground, and the problems of complex scenes, computing resources, hardware cost and the like are solved while naturalness of gesture interaction is improved.

Description

technical field [0001] The invention belongs to the technical field of gesture control, in particular to a space gesture control method based on deep learning model cascade. Background technique [0002] Gesture is one of the important ways of human-computer interaction. Compared with other interaction methods, gesture interaction has a more open scene and a higher degree of freedom. Traditional gesture recognition is mainly based on artificial design features for detection and recognition. The requirements are high, and gesture interaction often fails when encountering complex environments. At the same time, with the development of depth image technology, gesture interaction has gradually expanded from a two-dimensional plane to a three-dimensional space. The increase in spatial dimensions has greatly increased the computational cost of gesture interaction. [0003] In recent years, with the development of deep learning, gesture detection in complex scenes has lower and lo...

AI Technical Summary

Problems solved by technology

However, the complexity of the deep learning algorithm is relatively high, and a corresponding embedded computing platform is required to match it. In the application process, the current mainstream edge computing hardware includes GPU, SoC, FPGA, ASIC, etc.

Gesture interaction is the interaction method that is closest to the human life scene. In addition to performance, cost control must be considered when the application is implemented. The high-complexity deep learning algorithm has a high recognition accuracy in complex scenes, but with this At the same time, the greater the computing resources that need to be consumed, the higher the hardware cost required to carry the algorithm

Images