Distributed Teaching Video Processing System Based On Cloud Storage

With the development of the Internet,the demand for online teaching systems is very strong,and the processing of video teaching materials has become a more important research and application field.Teachers face huge challenges when processing 1080 P or 4K video materials;in addition,some teaching contents also have privacy protection issues.In the project of “Protection and Inheritance of Famous and Traditional Chinese Medicine” of Guangdong Provincial Hospital of Traditional Chinese Medicine undertaken by the laboratory,it not only has a high demand for high-performance real-time video processing,but also needs to protect the privacy of patients in teaching videos.Aiming at the problem of the above teaching video material processing,the thesis carried out research on related algorithms of “automatic recognition and mosaic of human eyes”,so that the patient privacy algorithm for protecting videos can be automatically processed in real time or near real time,and a cloud storage-based The distributed high-performance teaching video processing system,with the help of GPU cloud processing system,supports multiple teachers to use the cloud to transcode the teaching video content,adjust the resolution,etc.to meet different video transmissions during the online course production of the learning system Bandwidth requirements.The main contributions of this article are as follows:1.This paper studies and implements a complete "human eye automatic mosaic" to protect people's privacy system,which consists of three parts: face detection,face verification and human eye feature point detection.The face detection algorithm is mainly based on the improvement of the Face Boxes network structure.After the improvement,the average accuracy rate is increased from 86.4% to 90.4% at the cost of a slight increase in computing time.The face verification algorithm is improved based on Mobile Face Nets(mobile face network),so that the accuracy of face verification remains almost unchanged,the CPU's computing speed has increased by nearly 4 times,from 38 FPS to 132 FPS.Human eye feature point detection module We designed a 6-layer neural network to detect 18 human eye feature points from the human face.Finally,we composed three modules to form a complete human eye automatic mosaic system,which can achieve a speed of 94.3ms/frame on the CPU and 18.9ms/frame on the GPU(up to 25 frames per second)for 1024×1024 resolution video Real-time processing speed),and the accuracy rate of mosaic of patients in the diagnostic video of traditional Chinese medicine reached an average merge rate of 83.4%,which initially met the requirements.In addition,we proposed the idea of performing face verification experiments on mosaic faces in early days(basically similar to the concept of adversarial neural networks);2.This article completes a distributed teaching video processing system based on cloud storage that supports multi-user and multi-tasking.The system transfers local teaching video processing tasks to the cloud,and uses GPU hardware acceleration capabilities to achieve video cropping and conversion.Accelerate the functions of code,adjust resolution,adjust bit rate and mosaic with human eyes.Experiments show that the GPU-accelerated video processing system in single-task mode has increased the speed by at least 5 times compared to personal PCs.The processed materials can be directly transmitted to the teaching platform,which greatly improves the efficiency of teaching video processing and avoids Teacher groping on personal computer.The above algorithm and system have been completed.We applied for a soft cover and a patent for the system.The main work participated in the "Fifth Next Generation Internet Technology Innovation Competition" on December 8,2019 and won the national second prize,Ranked fourth in the country.At present,this system has been put into trial use and achieved good results.