当前位置:网站首页>Developing those things: how to solve the problem of long-time encoding and decoding of RK chip video processing?
Developing those things: how to solve the problem of long-time encoding and decoding of RK chip video processing?
2022-07-19 11:34:00 【TSINGSEE】
Streaming video live includes the following steps : collection —> Handle —> Coding and encapsulation —> Push stream to server —> Server stream distribution —> Player streaming .

In the process of streaming media processing and coding , There will be two playback modes: hard decoding and soft decoding . The difference between the two is , Hard decoding is to use the special decoding chip on the hardware device to speed up decoding . Generally speaking , The decoding speed of hard decoding is relatively fast , But there is also another problem , That's compatibility . As hard decoding is implemented by various hardware manufacturers , Uneven quality , So it is easy to have compatibility problems .

Soft decoding uses CPU The computing power to decode , If CPU The ability is weak , Then the decoding speed will be slow , And the equipment will also produce heating and other phenomena , But because of the unified algorithm , So the compatibility is good . For the same platform , The same hardware environment , Hardware coding is faster than software coding ,CPU Lower occupancy , So under the condition of hardware coding , Try to use hardware coding .

When testing products, technicians found , stay RK Used in Ruixin micro mpp When hardware encoding and decoding , Will decode the yuv Data is copied through memory function memcpy, It takes a long time to copy , about 80-100ms , This will cause the decoded data speed to fall behind , It will affect the playback effect .

In this case , We also provide a solution . The process here is 264->yuv->jpeg, for fear of yuv Data copying takes time , The new interface input parameter is H.264 data , Frame frequency , A reference to jpeg data . because yuv Direct use of data mpp Of memory , This operation reduces memory movement , Therefore, the encoding and decoding time is greatly reduced , Meet business needs .
As a family, it originated from EasyDarwin Open source framework for technology-based enterprises , In the development process , We are also happy to communicate code technology with all technology development enthusiasts 、 Share development experience , To provide you with some small reference .

TSINGSEE Qingxi video focuses on the research and development of video streaming media technology , In terms of video capabilities and services , Continuous combination AI、 big data 、 Edge of computing 、 Internet of things and other emerging technologies , Enrich platform function modules , Reach the diversified needs of users . At present, we have launched diversified video platforms for the market , Such as EasyCVR、EasyGBS、EasyDSS、EasyNVR etc. , It can realize the access of multiple types of equipment 、 Multi protocol support 、 Multi format video stream output 、 Multi terminal platform compatibility . Interested users can go to the demonstration platform for experience or deployment testing .
边栏推荐
- 开发那些事儿:如何解决RK芯片视频处理编解码耗时很长的问题?
- 微服务上线规范
- Category imbalance in classification tasks
- 学习笔记3--规划控制中的机器学习基本思想
- A curated list of awesome Qt and QML
- Tire Defect Detection Using Fully Convolutional Network-论文阅读笔记
- What should I do if I can't see any tiles on SAP Fiori launchpad?
- 夢想CMS 前臺搜索SQL注入
- The difference between CPU load and CPU utilization
- Discussion on Euler angle solution of rocket large maneuvering motion
猜你喜欢

Keras deep learning practice (14) -- r-cnn target detection from scratch

PowerCLI 脚本性能优化

Keras深度学习实战(14)——从零开始实现R-CNN目标检测

03-1、内联函数、auto关键字、typeid、nullptr

Leetcode 1328. Destroy palindrome string (yes, solved)

【无标题】cv 学习1转换

Today's sleep quality record 79 points

Will causal learning open the next generation of AI? Chapter 9 Yunji datacanvas officially released the open source project of ylarn causal learning

委派双亲之类加载器

466-82(3、146、215)
随机推荐
A curated list of awesome Qt and QML
Some methods of early MCU encryption [get data in the comment area]
Play with the one-stop scheme of cann target detection and recognition
Detailed explanation of multiple linear regression
Introduction to virtualization troubleshooting
翻墙后看什么?最热门的国外网站——翻墙网址导航
Leetcode 1310. 子数组异或查询
LeetCode 558. Intersection of quadtree
What happened to cinder in openstack-m
虚拟化排错概论
MySQL autoincrement ID, UUID and snowflake ID
Category imbalance in classification tasks
Leetcode 1304. 和为零的 N 个不同整数
Mysql 自增id、uuid与雪花id
Unity3d read mpu9250 example source code
Mysql索引的类型(单列索引、组合索引 btree索引 聚簇索引等)
Detailed explanation of MySQL show processlist
Evaluation method of machine learning model
Today's sleep quality record 79 points
mysql show processlist 详解