Giter VIP home page Giter VIP logo

esp32-cam-fpv's Introduction

esp32-cam-fpv

esp32 cam digital, low latency FPV

This project uses a modified esp-camera component running on an AI Thinker board to send low-latency mjpeg video to a Raspberry PI base station with 1-2 wifi cards using packet injection and monitor mode.

It uses FEC encoding (4/7 currently configured) with 1400 byte packets and it achieves quite good performance:

  • Up to 12Mbps video rate.
  • More than 90 FPS at 400x296 or lower with 20-50 ms latency.
  • 34-44 FPS at 800x600 or 640x480 with 50-80 ms latency.
  • 12 FPS at 1024x768 with >100 ms latency.
  • Air unit weighs in at 18g with camera, antenna and a 3D printed case.

It's based on an Ai Thinker board with an OV2640 camera board and the esp-camera component.

The data is received from the camera module as JPEG ar 20MHz I2S clock and passed directly to the wifi module and written to the SD card if the DVR is enabled. The ESP camera component was modified to send the data as it's received from the DMA instead of frame-by-frame basis. This decreases latency quite significantly (10-20 ms) and reduces the need to allocate full frames in PSRAM.

The wifi data is send using packet injection with configurable rate - from 2 MB CCK to 72MB MCS7 - and power.

The air unit can also record the video straight from the camera to a sd card. The format is a rudimentary MJPEG without any header so when playing back the FPS will be whatever your player will decide.
There is significant buffering when writing to SD (3MB at the moment) to work around the very regular slowdowns of sd cards.

The receiver is a Raspberry PI 4 with 2 wifi adapters in monitor mode (TL-WN722N). The adapters work as diversity receivers and the data is reconstructed from the FEC packets.

The JPEG decoding is done with turbojpeg to lower latency and - based on the resolution - can take between 1 and 7 milliseconds.
It's then uploaded to 3 separate textures as YUV and converted to RGB in a shader.

The link is bi-directional so the ground station can send data to the air unit. At the moment it sends camera and wifi configuration data but I plan to have a full bi-directional serial port for telemetry coming soon.
The back link uses 64byte packets with a 2/6 FEC encoding (so quite solid) at a low wifi rate (I think 1Mb).

This is very WIP at the moment, but it proves that the ESP32 can definitely stream video with very low latency.
I plan to use this is a long range micro quad.

Here is a video shot at 30 FPS at 800x600 (video converted from the source mjpeg):

session004_segment000.mp4

Building

esp32 firmware:

  • Uses esp-idf-v4.3-beta1, can probably work with newer idf versions. It needs to be properly installed (follow the instructions in the IDF documentation)
  • Only the Ai Thinker esp cam board tested
  • In the air_firmware, execute this: idf.py -p /dev/tty.usbserial1 flash monitor. Replace tty.usbserial1 with your serial port.
  • Make sure you place the board in flashing mode bu connecting IO0 to GND and resetting the board.
  • After compiling and resetting, you should get some stats per second, smth like this in the console:
    WLAN S: 695196, R: 350, E: 0, D: 0, % : 0 || FPS: 64, D: 401268 || D: 0, E: 0
    WLAN S: 761616, R: 420, E: 0, D: 0, % : 0 || FPS: 69, D: 443679 || D: 0, E: 0
    WLAN S: 763092, R: 420, E: 0, D: 0, % : 0 || FPS: 69, D: 449410 || D: 0, E: 0
    WLAN S: 764568, R: 420, E: 0, D: 0, % : 0 || FPS: 69, D: 450996 || D: 0, E: 0
    WLAN S: 761616, R: 350, E: 0, D: 0, % : 0 || FPS: 69, D: 449347 || D: 0, E: 0

Raspberry Pi ground station:

  • I use a Raspberry Pi 4 in a RasPad3 enclosure, but any HDMI display should work. Raspberry Pi 3 should work as well.
  • You need to use 2 TL-WN722N adapters connected to USB. Check the EZ-wifibroadcast wiki for more info about the hardware revision of these adapters and alternative adapters. Make sure you get the 2.4GHz ones, of course. NOTE: the adapters are critical, not all work in monitor mode!
  • If you only have one adapter or they are not called wlan1 & wlan2, check the main.cpp file and change the names and number there:
    rx_descriptor.interfaces = {"wlan1", "wlan2"};
    tx_descriptor.interface = "wlan1";
    Eventually this should be command line driven.
  • The UI uses ImGui and is touch driven - but mouse should work as well
  • Dependencies: sudo apt install libdrm-dev libgbm-dev libgles2-mesa-dev libpcap-dev libturbojpeg0-dev libts-dev libsdl2-dev libfreetype6-dev
  • In the gs folder, execute make -j4
  • Run sudo -E DISPLAY=:0 ./gs

The GS can run both with X11 and without. However, to run it without GS you need to compile SDL2 yourself to add support for kmsdrm: git clone https://github.com/libsdl-org/SDL.git
cd SDL
mkdir build
cd build
../configure --disable-video-rpi --enable-video-kmsdrm --enable-video-x11 --disable-video-opengl
make -j5
sudo make install

To run without X11 using the just compiled SDL2, do this:
sudo -E LD_LIBRARY_PATH=/usr/local/lib DISPLAY=:0 ./gs

Some other things that should improve latency:

  • Disable the compositor from raspi-config. This should increase FPS
  • Change from fake kms to real kms in config.txt: dtoverlay=vc4-fkms-v3d to dtoverlay=vc4-kms-v3d

VSync is disabled and on a PI4 you should get > 200FPS if everything went ok and you have configured the PI correctly.

esp32-cam-fpv's People

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.