DJI-Windows-SDK-UWP-Learning

---

Official Sample: https://github.com/dji-sdk/Windows-SDK

---

Prerequisite

To use DJI windows SDK:

• A PC or laptop with a Wi-Fi adapter(5GHz recommended)
• win7 or above
• vs 2017
• Mavic Air with latest firmware

* Setup Environment

• Install DJI Assistant 2
• Prepare Mavic Air
• Change to 2.4GHz
• Connect Mavic Air to DJi Assistant 2
• Setup Summary

---

Build And Run SDK Sample

There are two projects included in the solution:

• DJIWindowsSDKDemo：demonstrates the usage of SDK's interfaces
• DJIVideoParser：used to decode the video from the aircraft(FFmepg is used in the parser)

---

SDK Explanation

Since official version of this SDK hasn't been released now, what used here is an alpha version. It is still in vert early stage and not mant features are supported yet. It consists of four modules：SDKManager,VideoFeeder,VirtualRemoteController and ComponentManager

1. SDKManager Only entry of DJI Windows SDK. It contains the other modules.
2. VideoFeeder Used to receive the real-time video feed.不同的飛行器會返回不同的視頻流，當前的SDk只支持基礎的視頻流
3. VirtualRemoteController SDK和Mavica Air通過Wi-Fi連接，充當一個虛擬遙控器。該模塊包含一個SetJoyStickValue(float, float, float, float)介面以模擬操縱桿
4. ComponentManager SDK將飛行器抽象成多個部分，該模塊包含各個模塊的管理

---

註意事項

• 當SDKManager 激活的時候SDKManager返回的ComponentManager和其返回的句柄必須為non-null[沒有飛行器連接的時候句柄可用但會報錯]
• 執行結果並不是立即返回的，SDK和飛行器之間的信息交互是異步的

---

error occured when building,debuging...

1. 錯誤 C7510 “Callback”: 模板 從屬名稱的使用必須以“模板”為前綴

• C/C++ -> 語言 -> 符合模式：關閉/permissive-[取消禁用多語言擴展，本是為了適應跨平臺跨編譯器編譯Windows C++代碼做的檢查，但是攔住了WRL里的Callback末班函數的調用]

---

---

Deep-Diging into DJIVideoParser with ffmepg

深入理解視頻解析部分

1. djivideoparser.cpp：入口

• 初始化和銷毀
• SetWindow(int product_id, int product_type, int index, void* window)
• set_callback(int product_id, int index, std::function<void(uint8_t *data, int width, int height)> func)
• set_parser_data(int product_id, int index, const unsigned char *data, size_t data_length)

2. modulemediator.cpp:VideoParserMgr類型的共用內存區的管理

• 初始化和銷毀

3. videoparsermgr.cpp：管理各個設備傳回的視頻流，放在一個videoparser.VideoParser類型的共用內存區

• 初始化和銷毀
• AddDevice(int product_id)
• ParserData(int product_id, int component_index, const unsigned char* data, unsigned int len)
• RemoveDevice(int product_id)

4. videoparser.cpp：VideoParser類型的共用內存區處理數據的入口

• 初始化和銷毀 對Previewer、DjiCodec和VideoWrapper進行初始化
• ParserData(const unsigned char* buff, unsigned int size) 根據不同的碼長區分不同飛行器返回的數據流，然後調用VideoWrapper.PutToQueue()

5. VideoWrapper.cpp：封裝了視頻解析中的視頻幀回退清空暫停等操作

• s_rate=30 -> fps?
• 初始化和銷毀
• SetVideoFrameCallBack(std::function<void(uint8_t *data, int width, int height)> func)
• PauseParserThread(bool isPause)
• ClearFrame()
• PutToQueue(const uint8_t* buffer, int size, uint64_t pts) 根據pts區分數據流是實時視頻還是視頻幀，然後調用H264_Decoder.videoFrameParse()
• PutVideoToQueue(const uint8_t* buffer, int size, uint64_t pts)
• FramePacket(uint8_t* buff, int size, FrameType type, int width, int height)

6. h264_Decoder.cpp：解碼部分with ffmepg

• videoFrameParse(const uint8_t* buff, int video_size, FrameType type, uint64_t pts) 將數據流加入隊列，填充ffmepg

7. Queue.cpp：基本數據結構

8. Utils.cpp：工具

---

---

---

**From Official github samples:DJI SDK Sample for Windows 10 and Universal Windows Platform

This sample is a Universal Windows Platform* app that demonstrates DJI Windows SDK capabilities:

• Aircraft's camera video feed
• Aircraft control (Joysticks and Takeoff/Home functionality)
• Camera gimbal control
• Flight telemetry data events

The app leverages XAML-based UI and Windows ML, running inference (evaluation) on top of aircraft's video feed using a Windows ML model.

Windows ML

This sample uses a WinML model called InkShapes. It was made by Nikola Metulev and trained in Custom Vision on simple black-and-white hand-drawn pictures representing one of 21 categories, such as house, flower, stick figure, bike, and others.

With Custom Vision, you can train an AI model using your own images, then export it to different model formats including ONNX.

The model is not smart enough yet. House, flower, and stick figure are recognized quite well. You may help Nikola training the model - just by drawing shapes in Draw the shape! app.

Requirements:

1. Windows 10 April 2018 Update.
2. Windows 10 SDK 1803 (for April 2018 Update).
3. Visual Studio 2017 with Universal Windows Platform tools (including C++ tools for UWP), Desktop C++ Development and Desktop .NET development.
4. DJI SDK for Windows. All BUILD 2018 attendees should have a link to DJI Windows SDK alpha.

How to build

You need DJI Windows SDK.

Prepare the SDK:

DJI Windows SDK is not included in this repo. You need to obtain it from DJI, and copy 2 folders from the SDK to this project's folder.

1. Copy DJIWindowsSDK folder from the SDK's package to the root of this project.
2. Copy ThirdParties folder from the SDK's sample (DJIWindowsSDKDemo/ThirdParties) to the root of this project.

Note, these 2 folders are in .gitignore, so if you want to have them in your repo, please remove these 2 lines from .gitignore:

ThirdParties/
DJIWindowsSDK/ 

Build with Visual Studio 2017:

1. Once you copied all required dependencies, open WinDrone.sln in Visual Studio 2017
2. Right-click on DJIUWPDemo, click 'Set as StartUp Project'
3. Build, debug, deploy, enjoy!

*Platform notes

Current alpha version of DJI SDK only supports x86 architecture on Windows desktop. It has a few "classic" Win32 dependencies, so the app package requires Full Trust capability (using Desktop Bridge), and only runs on Windows 10 Desktop. Video decoding component doesn't leverage hardware acceleration yet.

The sample C# app uses the SDK via an additional DJIClient DLL and PInvoke calls.

Full Universal Windows Platform support and other improvements are coming later towards the release of DJI Windows SDK.