Coder Social home page Coder Social logo

limon93 / mitsubishi-m64282fp-dashcam Goto Github PK

View Code? Open in Web Editor NEW

This project forked from raphael-boichot/mitsubishi-m64282fp-dashcam

0.0 0.0 0.0 65.31 MB

A mobile recorder for the Mitsubishi M64282FP sensor of the Game Boy Camera

License: GNU General Public License v3.0

C++ 9.56% C 88.53% MATLAB 0.90% PowerShell 1.00%

mitsubishi-m64282fp-dashcam's Introduction

The DASHBOY CAMERA

A digital camera made from the Mitsubishi M64282FP artificial retina and a Raspberry Pi Pico.

The project fully emulates the sensor management strategy of the Game Boy Camera while completely bypassing the MAC-GBD mapper, which means you can record pixel perfect images without the limitation of storage space or the need for a Game Boy Printer emulator. The device can output dithered 2 bits per pixel images (like the Game Boy Camera) as well as smooth 8 bits per pixel images that take advantage of the full sensor bit-depth. It features time lapse recording, live recording, motion detection, High Dynamic Range, multi-exposure, night mode, Peak Focusing, fancy borders and even more. It's a fully open project so fork it and play with it !

The code was developped for Raspberry Pi Pico with the Earle F. Philhower RP2040 core. TFT display is driven with the Bodmer TFT_eSPI library. The code originates from an Arduino version that requires a PC. This project was preceded by a regrettable attempt to do the same with an ESP32 (to be honest, I did not know what I was doing at that time but the dev on ESP32 was a nightmare in comparison to Pi Pico). You may also check this similar project using a STM32 and a Holga camera.

Want to discuss with the author or other Game Boy Camera nerds ? Join the Game Boy Camera Club !

Picture of the device (example of PCB V1.0)

Dashboy Camera V1.0

Multi-systems installation guide

  • Just drop the compiled uf2 file to the board in mass storage media mode (Connect the Pico to USB port with the BOOTSEL button pushed and release).

Or if you like pain:

  • Install the last Arduino IDE

  • Install the Earle F. Philhower Raspberry Pi Pico Arduino core for Arduino IDE via the Arduino Board manager (see installation guide).

  • Install the Benoit Blanchon Arduino JSON library via the Arduino library manager.

  • Import or install the Bodmer TFT_eSPI library via the Arduino library manager.

  • Locate the TFT_eSPI library: \Arduino\libraries\TFT_eSPI folder in your Arduino libraries

  • copy the configuration file for the TFT display in this folder.

  • edit the User_Setup_Select.h and modify line 29:

    #include <Mitsubishi_M64282FP_dashcam_TFT_eSPI_setup.h> // Default setup is root library folder

  • Compile your code and flash the .uf2 to your board (Arduino IDE does that automatically once you've pressed the BOOTSEL button once).

Basic user manual

The Dashboy Camera uses a configuration file that must be placed at the root of SD card which has priority on the internal default parameters. The device cannot boot without a working sensor and a working SD card. Status of sensor, SD card and configuration file are indicated on the splashscreen.

  • Once the device is booted, it adapts the sensor exposure for 2-3 seconds then runs in Display mode. Display mode does nothing else than showing on display what the Mitsubishi sensor sees at the maximum allowed refresh rate. The green LED indicates the exposure time. It can go from approx 1 ms to 1 second depending on lighting conditions. The red LED indicates access to the SD card for recording (or error at boot).
  • There are three main modes: Regular Camera mode to take picture one by one like any camera, Motion sensor mode that triggers pictures according to image change and Timelapse mode to automatically take pictures with a delay inbetween. Shifting from one mode to the other is made by pushing the TIMELAPSE button. It rolls over a list of 6 exposure delays for time lapse and the two other modes. In the json configuration file, delays are entered in ms, a value of -1 indicates a return to regular camera mode, a value of -2 indicates a shift to Motion sensor mode. Keep the total length of the table at 8 entries !
  • In Timelapse mode, each recording session uses a different folder to store images. Timelapse mode has virtually no ending, it can go as long as the SD card is not full. The device enters a sleeping loop every 1/10 of the timelapse delay and wake up only briefly to accomodate exposure. The happy consequence is that electrical consumption falls to almost zero. The drawback is that the interface is less responsive. You typically have to push buttons for a certain time to get a response.
  • In Regular camera mode, pressing the button just records one image in the /camera folder. If the button stays pushed, images are recorded continuously with a debouncing delay of 150 ms. It acts more or less as a burst mode.
  • In Motion sensor mode, the DashBoy camera takes pictures automatically when it senses a motion. The threshold can be configured in the json configuration file.

To activate HDR mode or Dithering mode, simply push the corresponding pushbuttons.

HDR mode take several images from -1EV to +1EV and make an average of them before recording. This increases the dynamic of the sensor. The list of exposures can be modified in the config.json. Keep the total length of the table at 8 entries !

Dithering mode mimicks the dithering process of a Game Boy Camera with 4x4 derived Bayer matrices. Dithering matrices generated online can be copied in the config.json file.

Finally, you can lock exposure time from the current best one in any mode by pressing the LOCK_EX push button. It displays a green screen border when it is on.

It is mandatory to format the SD card in FAT32 and it is better to use the maximum sector size possible to speed up writing and avoid stalling. The access to the SD card is indeed the bottleneck in Recording mode.

The device outputs 1x BMP images natively but the projects comes with a bunch of tools to convert and upscale them and make videos from image series.

Advanced user manual

Additionally, you can address other cool features by entering them in the config.json:

  • Raw recording mode available in time lapse mode and motion sensor mode, records sensor data as a single file in the /TL or /MS folders respectively. It needs a separate tool for extracting an animation (provided here). The 16 first bytes are keywords, image sizes (width and height in pixels) and registers, the following are raw data for one image in reading order, easy to recover with any other tool ! This mode allows recording at a solid 4 fps for long time with good light conditions. It is disabled in Regular camera mode (command is just ignored).
  • Night Mode allows automatically downclocking the device in case the exposure registers reaches their theoretical maximal value (0xFFFF). This is usefull to do light painting from example, the sensor being natively not able to expose more than 1 second. Here there is no limit.
  • Fixed exposure mode allows bypassing the autoexposure algorithm and fixing a value, useful for astrophotography where auto-exposure performs poorly. In this case you have to enter the exposure time or delay (between 0x0030 - 0.768 ms and 0xFFFF - 1.044 second) and the clock divider (a multiplier for the exposure time, stay at 1 if you do not know what to do.)
  • Pretty border mode generates 160x144 images with a fancy border like the Game Boy Camera. You can make your own with the tools provided to generate C-files.
  • 2D enhancement mode enforces the 2D image enhancement processed by the sensor in Game Boy Camera mode & non dithering mode only.
  • Game Boy Camera mode drives the DashBoy Camera exactly as the Game Boy Camera does, using the same registers and the same dithering matrices in default constrast. Images made in this mode are pixel identical to images produces with a Game Boy Camera in the same lightning condition if dithering is activated. When set to 0, this triggers a simplier "recipe" with fixed registers and fixed dithering matrix.
  • Focus peaking allows embossing the area with the best exposure as focusing with such low definition can sometimes be tricky.

Some pictures made with the device

Dashboy Camera pictures

Pictures were taken with the original plastic lens, CCTV lenses with CS mount mod, CCTV fisheye with M12/CS adapters and a Tair-3s 300mm f/4.5 soviet telelens, sensor glued into a Zenit 12XP shell.

Building the device !

Required parts

The total cost (Game Boy Camera not included), considering that you start from scratch, is about 35โ‚ฌ (PCB included). It has been designed to use ultra-cheap and ultra-common parts only. Most of parts can be recycled from used electronics or Arduino related projects. If you're a bit into electronics, you probably yet have most of the parts somewhere on shelves. It is fully through-hole so it necessitates nothing but very basic skills in soldering.

So you will need:

The device is meant to be used with AA NiMH batteries. Lithium batteries are evil because they are unstable, not well recycled and non generic at all in terms of size in case of reuse. The device does not allow to recharge these batteries as NiMH requires very specific charging strategy. It can run with alkaline batteries too (even pretty discharged). You can of course alter the design to use it with Lithium batteries and embedded charger, it works perfectly too (my first design used a DD05CVSA charge discharge circuit). You can consult here the pinout of the prototype to adapt a lithium battery charger. Beware, the prototype pinout differs significantly from the current PCB.

PCB and connection with the sensor

PCB are available in the PCB folder of the project. To connect the sensor to the board, use a female JST connector with bended pins in order to inverse the gender of the JST cable, or simply order the breakout PCB. Be carefull with the polarity: both faces of the cables must be the same (camera sensor cable side and extension cable side, see following images). This situation corresponds also to the INPUT<->NORMAL OUT of the breakout board if you order the extension cables given in Required parts section.

Dashboy Camera connections

PCBs can be ordered at JLCPCB by simply uploading the gerber zip container to their site. JLCPCB has the advantage to be cheap and clean with the VAT for EU customers. Order HASL finish with 1.6 mm thickness for the two boards (default option). PCB designs and schematics can be edited with EasyDA.

Building instructions

  • Solder the Pi Pico first: it must be mandatorily soldered on the castellated holes, without pin headers or spacer, directly on the PCB
  • Solder the lower parts next: LEDs (RED and GREEN), resistors (RRED and RGREEN), level shifters, push buttons, JST connector, microswitches (MAIN and DISPLAY), capacitor (CAP1) and DC-DC converter. Levels shifters and DC-DC converters can be soldered with male pin headers, there is enough clearance for them.
  • Solder the display: ensure to have enough clearance to slide the SD card out of the display above the Pi Pico. Best is either to use female pin headers to secure the distance between display and main PCB or to solder the display with pins just showing out of the back PCB surface (pins are long enough).
  • Trim the remaining pin connectors as short as possible from the back side of PCB.
  • At this step, flash the Pi Pico and verify that the device boots and the display works. It should be stuck to an error screen with the red LED flashing, this is normal as the device needs SD card and a sensor to go further.
  • Solder the JST connectors to the tiny adapter board (INPUT and NORMAL OUT if you use a cable similar to the default sensor one linked in Parts) or just bend straight the pins of a female JST connector and plug it into the JST extension cable to inverse its gender.
  • Trim the AA battery holder ON/OFF switch and place it in ON position, solder the battery terminals.
  • Glue the Battery holder behind the pushbuttons (it will act as a hand grip) and the camera shell behind the display. Use double sided tape or dots of hotglue in order to allow reversing the mod and get back your precious camera shell if necessary.
  • Connect the sensor to the extension cable or adapter board (beware of the polarity), place fresh AA batteries and enjoy your Dashboy Camera !
  • If you want to use the time lapse or HDR features, use a tripod and hold the Dashboy with any universal phone holder by the battery holder.

Image of the populated PCB

Dashboy populated PCB

Some random informations for you Game Boy Camera nerd !

  • The sensor has intrinsically many visual artifacts (horizontal and vertical lines, image echoes at certain exposures, etc.). None of them is specifically due to the Raspberry Pi Pico. They were all observed with a real Game Boy Camera in the same lighting conditions.
  • The 16 bits exposure register range is normally limited to [0x0030-0xFFFF] despite the camera being technically able to cover [0x0001-0xFFFF]. This is advised in the sensor datasheet and the reason is that below 0x0030, the image artifacts becomes so intense that they interfere with any auto-exposure algorithm based on pixel intensity. Technically, this creates exposure jitters and unpleasant behavior. The Game Boy Camera mode have a low range of exposure extended to 0x0010 by playing on other registers at the same time but this however comes with a dark halo on the left side of image at very low exposure time (typically full sun in summer).
  • The strategy used in Game Boy Camera mode and the auto-exposure algorithm have been sourced from real data logging on a regular Game Boy Camera. Source files can be found in this other repository. Detailed explanations can be found here.
  • Close up images of the sensor and the MAC-GBD decapped can be found here. The sensor presents masked pixels which always return the reference voltage given by register V plus or minus a small bias (dark current). Register O is intended to cancel this bias. They occupy the lines 123 to 128 of a 128x128 image. Correcting this bias is not mandatory to get clean images as long as digital post-processing is available. In this regard, the Game Boy Camera uses registers to target a quite narrow voltage range as it has very limited post-processing capabilities (mainly handled by the MAC-GBD mapper which returns directly 2-bits dithered images).
  • The role of each registers is quite well understood now and explained here. The Mitsubishi M64282FP datasheet is notorious for being crap so many informations were assessed by extrapolating the datasheet of the M64283FP which is basically the same sensor with more features.
  • The Game Boy camera uses only 5 bytes of registers while sensor requires 8 bytes. the 3 missing register bytes (P, M and X) are hard coded into the MAC-GBD and send to the camera anyway. Their main purpose is to inject custom convolution kernels like border extraction, not in use by the camera.
  • The M64282FP sensor is known to have been massively produced for use in the Game Boy Camera. No known old new stock or other electronic device using it have been publically documented yet. On the other hand, the M64283FP was produced but no device using it is known for sure at the moment. Old new stocks sometimes appears. Both chips are strictly pinout and software back compatible. One line of Japanese mobile phone (the THZ43 Chiaro / LaPochee by Mitsubishi) has used one of these sensor but it's unclear which without any tear down (the LaPochee device beeing horribly expensive, any tear down is unlikely to happen in the future).
  • The flash ADC of the MAC-GBD has a voltage full scale of 0<->3.3 Volts (even probably 3.36V, seing the effect of dithering matrices on Rapsberry Pi Pico raw data). Maximum voltage reached in normal use is about 3.05 volts. It can go until 3.18 volts during camera boot (may this be intentionnal or not).
  • The Game Boy Camera has a calibration procedure to adapt for each sensor which effect is not at all understood at the moment (it changes some sram values but the role of these data is not documented yet). It probably deduces a bias to apply in order to have consistent images with different sensors. Other say, the results you will get with the DashBoy Camera project may vary with the sensor used as I calibrate nothing on my side.
  • The Game Boy Camera has two PCB revisions for its sensors. It is unclear wether it has an effect on images other than subjective or if the board revision is associated with a sensor revision. Veteran users of the camera admit that early sensors (first PCB sensor version on Japanese cameras mainly) has more vertical amplification artifacts but there is no clear separation between revisions. Cleaning the lens generally cancels the differences.
  • If you use a lens mod with the DashBoy camera, how dirty will look your sensor just depends on the lens aperture you use. The wider the opening, the less artifacts on the image. These speckles due to dust or surface roughness can barely be seens with a regular Game Boy Camera due to dithering but becomes very obvious in 8 bits mode where the image is much smoother.

Some thought for free

According to an internal Mitsubishi source, the use of the M64282FP artificial retina for road safety dashcam application was assessed in 1999. They recommend using the MAX153 flash ADC to convert analog signal fast enough for a live (5 fps !) rendering and recording of images. The MAC-GBD, mapper of the Game Boy Camera, embeds a flash ADC too on its chip. The probability is thin but not zero that the MAC-GBD flash ADC would be simply a MAX153 adapted for this custom mapper. A copy of this experiment is proposed here. And yes it works !

Kind warning

The code and current design come as it. If you're not happy with the current hardware, the PCB EasyDA design or the Arduino IDE, create your own, the licence allows it ! Push request with tested and working improvements are of course still welcomed.

Acknowledgments

Developpement steps

Dashboy Camera prototype

mitsubishi-m64282fp-dashcam's People

Contributors

raphael-boichot avatar

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.