Face image tracking method, device, computer equipment and storage medium

Abstract

The embodiment of the present invention discloses a face image tracking method, device, computer equipment and storage medium, comprising the following steps: acquiring the position information and offset parameters of the face image in the target image; The position information is corrected; the corrected position information is defined as the change position of the face image in the target image of the next frame. The convolutional neural network model trained to the convergence state can infer the accurate offset parameters of the face image from the current target image to the next frame of the target image according to the position information and characteristics of the face image in the current target image. Therefore, The position information of the face image in the next frame of the target image can be deduced after the current target image is acquired. Through the above method, it is possible to solve the tracking problem of lost images when the shooting and tracking processing speeds are not synchronized during the face image tracking process. At the same time, the requirements on the processing speed of the tracking system are greatly reduced.

Description

technical field [0001] Embodiments of the present invention relate to the field of image processing, in particular to a face image tracking method, device, computer equipment and storage medium. Background technique [0002] Face tracking is to continue to capture information such as the position and size of the face in subsequent frames on the premise of detecting the face, including face recognition and face tracking technology. In customs, airports, banks, video conferences, etc., it is necessary to track specific face targets. [0003] In the face key point technology in the prior art, first use the full convolutional network to extract the features of the face in the previous frame, and then estimate the position of the face in the current frame according to the position of the face in the previous frame, and then use the same The fully convolutional network extracts the features of the estimated face position in the current frame, and then compares the face features o...

AI Technical Summary

Problems solved by technology

[0004] The inventors of the present invention found in research that the frame rate of video in the prior art is between 25 and 30 frames per second, and the detection speed of a single process of the face detection algorithm is about 10 frames per second, while the prior art When tracking, it is necessary to calculate the features of the face and its nearby positions in the front and rear frames. Due to the large amount of calculation in the feature extraction process, it is difficult to achieve real-time purposes, so the response speed is slow and the target face image is easy to lose

Images