Face deduplication method based on deep quadratic tree in video monitoring

Abstract

The invention discloses a face deduplication method based on a deep quadratic tree in video monitoring. The method comprises the steps of (1) a face detection part including the steps of preparing positive and negative face samples, performing face detection on moving pedestrians in a monitoring video by using a trained deep quadratic tree model, and obtaining a face position, face confidence, face sharpness and the resolution of a face image, (2) face tracking part including the steps of performing face tracking according to an obtained face position of an initial frame and obtaining multiplefaces of the same person, and (3) face deduplication including the steps of determining the image quality of each face by using a face quality evaluation method for a face sub set of the same personand then selecting a face with best quality according to an evaluation value. A face quality evaluation score is obtained by weighting three indicators including face confidence, the face sharpness and the resolution of a face image. According to the method, the accuracy of face recognition can be effectively improved.

Description

technical field [0001] The invention relates to the fields of computer vision and artificial intelligence, and belongs to the technical field of face detection, tracking, and image quality evaluation in surveillance videos, and is specifically designed as a method for deduplicating faces in surveillance videos. Background technique [0002] In recent years, biometric recognition technology based on face recognition has been widely used in daily life, especially in the fields of finance and security. Although the recognition technology has developed rapidly, the face recognition effect in actual monitoring scenarios has been lacking. One of the very important factors is that the image quality used for face recognition is low, mainly reflected in blurred images and low resolution. Therefore, using face images with clear images, high resolution, and correct postures for face recognition can effectively improve the recognition accuracy. In order to achieve this goal, it is nec...

AI Technical Summary

Problems solved by technology

[0003] In view of the existence of (1) the method with high detection rate is slow, it is difficult to apply in real time to the video surveillance scene (2) the method that can be applied in real time, the detection rate is low and the false detection rate is high in the complex scene in real life , The robustness against occlusion and illumination changes is not strong (3) The direct information interaction utilization of several modules of face detection, face tracking, and face quality evaluation is not enough (4) The evaluation of face clarity takes a long time and the evaluation is not accurate And other issues

Images