hx-esp32-cam-fpv

July 29, 2026 · View on GitHub

Open source digital FPV system based on esp32cam https://romanlut.github.io/hx-esp32-cam-fpv/

Features:

esp32/esp32s3 + ov2640: 640x360 30fps, 640x480 30fps, 800x456 30fps, 800x600 30fps, 1024x576 12fps
esp32/esp32s3 + ov2640 with sensor overclocking: 640x360 40fps, 640x480 40fps, 800x456 40fps
esp32s3/esp32c5 + ov5640: 640x360 30/50fps, 640x480 30/40fps, 800x456 30/50fps, 1024x576 30fps
HQ DVR Mode: 1280x720 30fps (esp32s3 + ov5640) or 1280x720 12fps(esp32/esp32s3 + ov2640) recoding with maximum possible quality on Air, low FPS transmission to the ground
up to 1km at 24Mbps (actual transfer rate is ~8-10Mbps total with FEC 6/12), 600m at 36Mbps (actual transfer rate is ~10-12Mbps total with FEC 6/12) (line of sight)
latency 90-110ms
bidirectional stream for RC and telemetry 115200 Kbps (for Mavlink etc.)
Displayport MSP OSD
on-board and groundstation video recording

See a list of all supported sensor/boards combinations below.

Air unit variants (VTX):

esp32s3sense with ov5640 camera (recommended)
esp32s3sense with ov2640 camera
esp32cam with ov2640 camera
esp32c5 with ov5640 camera (experimental)
esp32/esp32s3/esp32c5 with ov3660 camera

Ground station variants (VRX):

Radxa Zero 3W/3E with rtl8812au wifi card(s) (recommended)
Raspberry Pi Zero 2W ... Raspberry Pi 4B with rtl8812au or AR9271 wifi card(s)*
Runcam WiFiLink VRX - 5.8GHz only (experimental)
Android Phone or Tablet with rtl8812au USB adapter
Oculus Quest 2/3 rtl8812au USB adapter
GS Software also can be run on x86_64 notebook on Ubuntu or Fedora Linux

Recommended hardware

Although the project can be compiled for various platforms, the recommended hardware for optimal performance is:

VTX: Seed Studio XIAO ESP32S3 Sense with ov5640 camera, M12 120° lens, 2dBi dipole
VRX: Radxa Zero 3W with dual rtl8812au wifi cards, 4x5dBi dipoles

Theory

Wi-Fi bandwidth is insufficient for uncompressed video streaming, and the ESP32 lacks the processing power for real-time video encoding. Fortunately, OV* camera modules can output frames as JPEG images. This project utilizes that feature to stream MJPEG video.

The camera continuously scans the image sensor and encodes the data row-by-row into JPEG format. These JPEG frames are transmitted to the ESP32 over I2S at 10 MHz (or 20 MHz for ESP32-S3 with OV5640), where they are optionally written to an SD card (if DVR mode is enabled), forward error correction (FEC) encoded, and streamed over Wi-Fi.

To reduce latency and memory usage, the esp32-camera component (https://github.com/espressif/esp32-camera) was modified to stream data directly from DMA as it arrives, rather than waiting for full frames. This cuts latency by 10–20 ms and minimizes PSRAM usage. As the camera captures the image, data is already in transmission to the Ground Station (GS).

Wi-Fi packets are transmitted using packet injection, a feature supported by the ESP32. FEC (forward error correction) ensures that the GS can recover from packet loss without needing acknowledgments or retransmissions.

The air unit can also record video directly to an SD card. To compensate for the inconsistent write speeds of SD cards, a 3 MB buffer is used. The recorded video is identical in quality to the streamed version.

JPEG image sizes vary depending on scene complexity. Adaptive JPEG compression dynamically adjusts the quality level to match the available bandwidth and maintain smooth streaming.

The Ground Station can be a Radxa Zero 3W, Raspberry Pi Zero 2W, up to Raspberry Pi 4, equipped with Wi-Fi cards in monitor mode such as Realtek 8812AU (recommended), Realtek 8821AU or AR9271. Dual Wi-Fi cards can be used for diversity reception.

Since release 0.6.3, Android phone and Oculus Qeust 2/3 can be used for Ground station. Realtek 8812AU or Realtek 8821AU wifi card are still required for Raw Broadcast mode. In APFPV mode, built-in wifi can be used.

8812au with a low-noise amplifier (LNA) is preferred; however, a high-power amplifier (PA) is not critical since the range is primarily limited by the ESP32's maximum transmit power of 100 mW (20 dBm).

JPEG decoding on the GS is handled by TurboJPEG, achieving decoding times between 1–7 ms depending on resolution. Frames are uploaded as textures and rendered to screen.

On-screen display (OSD) elements are drawn using OpenGL ES.

The link is bi-directional, allowing the Ground Station to send data back to the air unit. Currently, this is used for camera and Wi-Fi configuration, as well as bi-directional telemetry streaming (also FEC encoded).

Here’s an example video captured at 30 FPS, resolution 800×456, using an OV2640 sensor with a 120° M8 lens:

https://github.com/RomanLut/esp32-cam-fpv/assets/11955117/970a7ee9-467e-46fb-91a6-caa74871fc3b

Is it worth building?

Set your expectations low. This system begins with a very basic camera — the OV5640 — comparable to smartphone cameras from around 2005. Using such a sensor means accepting several limitations:

Noisy image
Poor brightness/contrast handling, especially in backlit scenes
Distorted and inaccurate colors
Weak low-light performance
Vignetting from cheap lenses
Blurry corners due to poor focus
High JPEG compression artifacts

In addition, the ESP32 lacks hardware video encoding capabilities. As a result, the video stream is transmitted as a series of JPEG images — an inefficient method that wastes bandwidth and reduces potential image quality.

Given its low resolution, esp32-cam-fpv should be compared to cheap analog 5.8GHz AIO FPV cameras, not modern digital systems.

What you gain compared to analog AIO systems? For roughly the same price, hx-esp32-cam-fpv offers:

Simultaneous video recording on both air unit and ground station
Digital OSD
Mavlink telemetry and RC control
Telemetry logging
Clean digital image with no analog interference

Key Drawbacks:

Blocky JPEG compression
No Wide Dynamic Range (WDR)
Distorted color reproduction
Poor low-light performance (especially with OV2640)
Inconsistent quality between camera units
Occasional frame drops

hx-esp32-cam-fpv is clearly outclassed by all commercial digital FPV systems in terms of image quality.

However, compared to other open-source digital FPV solutions like OpenHD, RubyFPV, or OpenIPC, it offers:

low cost air unit
Very compact size air unit (XIAO ESP32-S3 Sense)
Low power usage (under 300mA at 5V)
The same ground station hardware used for OpenHD/RubyFPV/OpenIPC can be reused — just swap the SD card.

https://github.com/RomanLut/hx-esp32-cam-fpv/assets/11955117/3abe7b94-f14d-45f1-8d33-997f12b7d9aa

Modes of operation: Raw Broadcast / APFPV

The Air Unit can transmit video either in Raw Broadcast mode or in APFPV mode.

In Raw Broadcast mode, packets are transmitted using packet injection over a connectionless link.

In APFPV mode, the camera creates a Wi-Fi access point and the GS connects to it.

In both modes, only one GS can be connected to the Air Unit at the same time.

Raw Broadcast is the primary and recommended mode of operation. APFPV mode is not recommended for general use. APFPV requires establishing and maintaining a Wi-Fi connection to the access point, which can be unreliable or may fail completely. At long distances or with weak signal strength, this can result in losing the video feed for an extended period of time on Android devices, and especially Oculus Quest.

In general, APFPV mode is recommended only for ground vehicles.

Air Unit can be flashed with either Raw Broadcast or APFPV firmware. In fact, the firmware itself is the same - the only difference is the initial mode setting. When switching between firmware types, make sure to fully erase the flash before reflashing. Otherwise, the mode may not change.

The Air Unit mode can be changed either from the GS menu or from the camera web interface

Note

Raw Broadcast mode works only with compatible RTL88XX wifi card.

Note

If the mode is changed from APFPV to Raw Broadcast using the GS menu, reconnecting to the Air Unit without an RTL8812AU adapter will no longer be possible. In this case, you must either reflash the Air Unit firmware or switch the mode back using the camera web interface.

https://github.com/RomanLut/hx-esp32-cam-fpv/assets/11955117/cbc4af6c-e31f-45cf-9bb4-2e1dd850a5d8

Building

Note

Please use release branch (it contains lastest release). master branch can be unstable.

Hint: For quick start, you can use Android GS and esp32cam in APFPV mode.

Air Unit

Air Unit Hardware	OV2640	OV3660	OV5640	2.4GHz	5.8GHz	OTA Update	USB Disk	Wi-Fi File Server	DVR	I @ 5V	Guide
ESP32-CAM	+	+		+		+		+	+	<300mA	Building guide
ESP32-S3 Sense	+	+	+	+		+	+	+	+	<300mA	Building guide
ESP32-C5		+	+	+	+	+		+	1/4 speed¹	<300mA	Building guide

¹ every 4th frame is recorded.

Air Unit FPS

Air Unit + Sensor	640x360	640x480	800x456	800x600	1024x576	1280x720
ESP32 + OV2640	30/40	30/40	30/40	30	12	12
ESP32 + OV3660	30	30	30	30	22.5	18.4
ESP32-S3 + OV2640	30/40	30/40	30/40	30	12	12
ESP32-S3 + OV3660	30/50	30/40	30/50	30/40	30	29
ESP32-S3 + OV5640	30/50	30/40	30/50	30	30	30
ESP32-C5 + OV3660	30/50	30/40	30/50	30/40	30	29
ESP32-C5 + OV5640	30/50	30/40	30/50	30	30	30

Ground Station

GS Hardware Variant	2.4GHz	5.8GHz	Building Guide
Radxa Zero 3W/3E	+	+	Building guide
Raspberry Pi Zero 2W	+	+	Building guide
Raspberry Pi 2/3/4	+	+	Building guide
Android	+	+	Building guide
Oculus Quest	+	+	Building guide
Runcam WiFiLink VRX		+	Building guide
Eachine Sphere Link	+	+	Building guide
Emax Wyvern Link VRX	+	+	Building guide
Ubuntu	+	+	Building guide
Fedora Linux Workstation	+	+	Building guide

Radxa/RPI/Linux supported network cards

Network card	2.4GHz	5.8GHz
RTL8812AU	+	+
RTL8811AU	+	+
RTL8812EU	?	+
AR9271	+	+

? The chipset itself (e.g. RTL8822BU, RTL8812BU, RTL8821CU) supports both 2.4 GHz and 5 GHz, but the finished hardware may not implement both bands.

Android/Oculus Quest supported network cards

This project uses OpenIPC devourer for Android/Oculus Quest Raw Broadcast Wi-Fi adapter support.

Tested cards:

Network card	2.4GHz	5.8GHz
RTL8812AU	+	+
RTL8821AU	+	+
RTL8822BU	?	+
RTL8812EU	?	+

The support of untested cards are not mature in devourer. There are bugs. Please use tested cards.

Untested cards:

Network card	2.4GHz	5.8GHz
RTL8811AU	+	+
RTL8814AU	+	+
RTL8812BU	?	+
RTL8811CU	?	+
RTL8821CU	?	+
RTL8812CU	?	+
RTL8822CU	?	+
RTL8822EU	?	+

? The chipset itself (e.g. RTL8822BU, RTL8812BU, RTL8821CU) supports both 2.4 GHz and 5 GHz, but the finished hardware may not implement both bands.

Features

Displayport MSP OSD

Bidirectional serial connection

OSD Menu

Camera web interface and OTA update

VR Mode

Latency

USB Mass Storage mode (esp32s3)

Other modes of operation

Wifi channels scanning

Test mode

Recordings playaback

Design considerations

Image resolution selection

Development

Development guide

Original project

Unsuccessfull attempts

References

esp32c5-airunit-nano https://github.com/us3r-d0e5nt-3x1st/esp32c5-airunit-nano

ESP32-C5-CAM_Hardware https://github.com/HaqqScripter/ESP32-C5-CAM_Hardware

Getting Started with Seeed Studio XIAO ESP32S3 (Sense) https://wiki.seeedstudio.com/xiao_esp32s3_getting_started/

FAQ

https://github.com/user-attachments/assets/b0c2f0b5-2106-4702-b434-837e8ce5914b