Face detection de-duplication method based on limited buffer of embedded device

Abstract

The invention provides a face detection de-duplication method based on limited buffer of an embedded device. The method comprises the steps that video frames are acquired from a video stream; an algorithm is processed; an algorithm output result is processed; a video frame buffer queue is processed; a disappearing face is processed, and a face image is encoded; and encoded data are acquired and provided for a subsequent server. Algorithm processing needs to carry out face detection, tracking, ID labeling and comprehensive scoring on the video frames. Subsequent de-duplication is based on a face ID and a scoring value. Algorithm output result processing continuously updates data with the face ID as an index. The information about the best face is recorded. After algorithm detecting is carried out, the video frames are buffered. The video buffer queue simply buffers the image with the reference count greater than 0. When the buffer queue is full, a face referenced in the video frame withthe smallest reference count is pre-coded and saved, and then the video frame is deleted to save buffer. The step is a key step of the method provided by the invention.

Description

technical field [0001] The invention belongs to the field of practical application of face detection, and relates to deep learning face detection technology, embedded device picture cache management technology, embedded device cache application optimization and other technologies. Background technique [0002] Face detection and recognition technology is a research hotspot in artificial intelligence and pattern recognition today. It has a wide range of applications in many fields such as security protection, financial applications, and smart people's livelihood. The main function of face detection is to provide services for subsequent recognition. [0003] With the development of technology, the application of face detection and recognition in public security and other fields is gradually expanding. Correct and efficient recognition will bring twice the result with half the effort to security prevention and public security criminal investigation, which will help liberate a lo...

AI Technical Summary

Problems solved by technology

The existing embedded face detection equipment, due to the limited memory, and the original video frame pictures occupy a relatively large memory space, the number of cached original video frames is limited, especially in the case of many targets, each cached video frame has a For useful face pictures, the simple filtering strategy cannot use the limited cache to select the optimal face picture. When the buffer is about to overflow, the strategy of grabbing relatively good pictures can only be passively adopted. When there are better face pictures, it is necessary to continue to capture pictures, so there are problems of multiple captures and poor quality of some face pictures

Images