Face key point calibration compensation method and device for end-side device

Abstract

The application discloses a face key point calibration compensation method for end-side equipment, comprising the following steps: reading a current frame image in a face video stream; inputting the current frame image into a deep learning model and receiving a reference key point set output by the deep learning model; reading a historical key frame in a key frame database and judging whether the historical key frame exists in the key frame database; if not, marking the positions of the reference key point set in the current frame image and outputting and displaying the marked current frame image; if yes, performing position compensation on the reference key point set according to key point calibration data of all historical key frames to obtain a final key point set, marking the positions of the final key point set in the current frame image, and then outputting and displaying the marked current frame image. The face key point calibration compensation method can improve the detection accuracy of face key points without the need of retraining the deep learning model. The application further discloses a device based on the face key point calibration compensation method.

Images