Method and system for acquiring the line of sight direction of a human eye by a single camera

Abstract

The invention relates to a method and system for acquiring the line of sight direction of a human eye by a single camera. The method comprises the following steps: S1, acquiring human face images of various angles by using an image data acquisition device, and using the image data acquisition device to acquire the data for neural network depth learning and training; Step S2, at least calculating the coordinates of the position of the human eye in the image acquired in step S1 and the coordinates of the position of the gaze position of the corresponding line of sight; Step S3: the human face images of each angle in the step S1 include the human eye position images, and the gaze position coordinates are input into the neural network for machine learning; when the human face image is inputted, the line of sight can be correctly outputted through the training; Step S4, collecting the human face image when the driver is driven by a real-time acquisition camera, and inputting the human faceimage into the trained neural network, wherein the neural network outputs the real-time correspond line-of-sight direction of the human face image. The method adopts the improved image data training method, coordinate system transformation method and the corresponding neural network model. The visual line of sight calculated by the network model has high precision, which meets the needs of variousautomatic driving/assistant driving.

Description

technical field [0001] This application relates to the field of image data acquisition for artificial intelligence. It specifically relates to the field of face image data collection. Background technique [0002] The determination of the direction of the human eye's line of sight includes the determination of the position of the human eye and the position of the gaze point of the human eye. After the above two points are determined in the real three-dimensional space, the line connecting the two positions is the direction of the human eye's line of sight. For the determination of the position of human eyes, what the prior art used is a multi-camera acquisition system, which uses multiple cameras to obtain human face images, but this system uses multiple cameras, which increases the complexity of the system, such as prior art: CN106840112A and CN103871045A , using multiple CMOS cameras to obtain the human eye position coordinates. [0003] There are also acquisition system...

AI Technical Summary

Problems solved by technology

[0006] In view of this, the present application provides a device and method for collecting the direction of sight of the human eye with a single camera, so as to solve the technical problem that the recognition error of the direction of sight of the human eye is large and inaccurate in the prior art

Images