Category Archives: Programming

Jetson Nano Install Notes

The Jetson Nano Developer kit – B01 is a small computer comprising of an NVIDA Maxwell GPU, Quad-Core ARM Cortex-A57 Processor and 4GB of Memory along with four USB 3 ports, Gigabit Ethernet, HDMI and Display Port output, main storage is on a MicroSD card and there is a variety of selection of expansion available via GPIO, I2C and UART. On the software side NVIDIA provide their JetPack SDK – a customised version of Ubuntu. This development kit has been produced to provide an entry point into Machine Learning, for which I will be using Python programming language. I got my board from Pimoroni

Jetson Nano
Jetson Nano
GPU 128-core Maxwell
CPU Quad-core ARM A57 @ 1.43 GHz
Memory 4 GB 64-bit LPDDR4 25.6 GB/s
Storage microSD (not included)
Video Encode 4K @ 30 | 4x 1080p @ 30 | 9x 720p @ 30 (H.264/H.265)
Video Decode 4K @ 60 | 2x 4K @ 30 | 8x 1080p @ 30 | 18x 720p @ 30 (H.264/H.265)
Camera 2x MIPI CSI-2 DPHY lanes
Connectivity Gigabit Ethernet, M.2 Key E
Display HDMI and display port
USB 4x USB 3.0, USB 2.0 Micro-B
Others GPIO, I2C, I2S, SPI, UART
Mechanical 69 mm x 45 mm, 260-pin edge connector

These notes cover my process of setting one up and links to the documentation, it not intended to repeat those install guides but to provide an install sequence and any additional commentary as needed. I’m going to assume you have a little experience of using the terminal and am familiar with using the bash command line – I’ve no idea how this would be done through the GUI.

jetson nano ports
Jetson Nano ports [source: NVIDIA]

Initial Startup

I followed the instructions for downloading and installing JetPack 4.4 on https://nvidia.com/jetsonnano-start I used a 64GB Class 10, UHS-I, U3, V30 SanDisk card. I formatted the card in a camera before using balenaEtcher to write the JetPack SDK image, this creates a partition of about 16GB on the card formatted to ext4, during installation the volume is resized to fill the card.

jetson nano connections
Jetson Nano connections [source: NVIDIA]

Despite using a good quality USB power supply with an output of 3 Amps at 5 Volts into the Micro USB port the computer would only boot long enough for the NVIDIA logo to appear on screen but after a few seconds the green power LED would go out and it would be off, the same happened when I tried a variety of USB power supplies used. I got round the problem buy using a 5 Amp power supply connected to the barrel jack J25 on the left (centre pin positive) and connecting the jumper J48 located just behind this connector.

Customising the Setup

There are a couple of thigs to do, get a network volume mounted and set the default version of python.

For the network share install samba, some network utilities and the nano text editor:

create a text file: sudo nano /etc/samaba/videoserver with the following:

And set the permissions sudo chmod 600 /etc/samba/videoserver. In this example I have a network share on my server; 192.168.1.30, called video. Create a mount point for the share: sudo mkdir /mnt/video now you need to edit fstab, sudo nano /etc/fstab and add your network connection to the end:

Reload fstab with sudo mount -a and check for any errors. Because of the way that Jetpack boots it does not appear to wait for the network so the share needs to be set to automount and this causes it to only appear in drive listings when accessed. Further reading can be found in this excellent guide to fstab: https://wiki.archlinux.org/index.php/fstab.

Jetpack 4.4 comes with two versions of Python, 2.7 and 3.6, I want it to default to 3.6 and while this is rather out of date I don’t want to go down the hole of upgrading just yet, you will also need to install pip and set pip3 as the default too.

and test its worked:

I did get an error later on, a crash was reported on the desktop when an occasional python 2 script ran. I fixed the error in /usr/sbin/l4t_payload_updater_t210 by changing the first line of the file from !#/user/bin/python to !#/user/bin/python2

Post Install Problems

A recent update occured, so I did the usual sudo apt get update && sudo apt get upgrade but one of the files gave a script error, this turned out to be with nvidia-l4t-bootloader, like so:

after some head scratching, it turns out this script is python2 only. I reverted back to the python 2.7 as the default, and tried again:

Afterwards I set the default back to python 3.6 again.

Setup the Machine Learning

There are three Machine Learning packages, OpenCV for Computer Vison, Tensorflow for machine learning models, and TensorRT – accelerated deep learning networks for image recognition.

OpenCV is already installed in Python:

Tensorflow installation can be found here: https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/ I found this to be straight forward but it did keep the computer busy for a while.

TensorRT can be found at: https://github.com/dusty-nv/jetson-inference/blob/master/docs/building-repo-2.md I used the instructions in the Quick Reference section and installed the default models without any errors, I have not yet had the opportunity to test it.

Further Reading

Extracting GPS data from the GoPro 7

While extracting the telemetry data from the GoPro is reasonably well documented I have found some gaps for getting the extracting utilities installed and when extracting and combining data from multiple files. These notes are for a Debian/Ubuntu installation in a BASH Shell.

Installing the gopro-utils

As I couldn’t find any straightforward instructions for installation, I’ll be going through everything I needed to do to get it working, you may have some of these packages installed already.

Now to get the gopro-utils and install them, I’m placing the source files into my Downloads directory, as well as the GPS data extractor we’ll be adding the other telemetry tools too, this is all a bit long winded.

Extracting the Data

You will need to find which stream in the video recording the data has been saved to, to find this use ffprobe to examine the recording and look for the stream that contains GoPro MET, for example:

You can see that what we are wanting is on stream 3, as far as I can tell this stays the same every time, I don’t know if it is different for other GoPro models.

This bash script extracts the GPS data in GPX format from all the GoPro GX recordings in the directory, other options have been commented out, if you are using Garmin VIRB edit there is also an option for use with that. The script creates two files, one that contains the raw data and another with the desired GPS data, the GPS output file has the same name as the recording, but in lowercase with a .gpx extension.

Merging GPX files

As the GoPro splits recordings into 4GB blocks, when extracting you will get a single GPX file for each recording. Many pages found by Google say that to create a single track from these all you need to do is append the files into one big file. This is wrong, what you end up with is a single file with many short tracks, when what you are after is one long track covering the entire journey. This bash script uses gpsbabel to create single merged file from the extracted GPX data, it creates a file called “gpsoutput.gpx”.

The next stage will be to write a script that combines all these and completes the job in one easy process.

Links and Sources

The Fridge Door is Open

My fridge door tends to rebound when closed staying open a smidgen and letting all the cold out. Rather than just checking that the door is properly shut, I thought it about time to have a microcontroller make a noise when the door has been left open too long.

This circuit uses an Arduino compatible Teensy LC for all the work, it has a phototransistor to sense the state of the fridge door light, a couple of LED’s one to indicate the power and another that comes on when the door is open. There is also Piezo buzzer to make an annoying noise after forty five seconds of door open time. The unit runs of a 3.7v rechargeable Lithium-ion battery and I have added a recharging circuit that takes power via the Teensy’s 5v USB port.

Circuitry

Note: These diagrams show 3.7v as the supply voltage. The Teensy LC can only tolerate a maximum 3.3V on the data pins, so these circuits are driven from the 3V output on the Teensy. They will all work without modification on the 5v Arduino Uno.

I have used a phototransistor to detect the fridge door light, there are two variants of this circuit light activated or dark activated, the 100k resistor can be replaced with a 100k variable if you need to adjust the sensitivity, the 330k resistor provides a weak pull-down on the output. The phototransistor is being used in switch mode to provide a logical output (rather than active mode which provides an output relative to the amount of light), so the output is connected to a digital input on the Arduino. The BC547 transistor is half of the darlington pair to provide extra gain on the output.

Dark Activated Switch
Light Activated Switch

I chose the light activated switch, either will do but will provide different logical outputs to your controller. The circuit is enclosed in a small box inside the fridge and connected by ribbon cable to the controller, the ribbon cable is flat and does not upset the magnetic ‘seal’ on the fridge door.

Component Connections

To make some noise I used a piezo buzzer from an old computer, this is driven through a transistor as the Teensy does not provide enough current to drive it directly.

Piezo buzzer

There is also a push button to provide a reset function if the buzzer is sounding while the door is open.

Push Button

I have also added two LED’s, one to show power and anther that illuminates when the door is opened.

Power and door LED’s

The final circuit if for recharging the battery, it connects to the 5V connection on the Teensy LC so charges the battery when the USB connection is in use. This has been copied from the MCP3831T datasheet.

Battery Charger

Software

This uses an interrupt to listen for the light sensor, when the state changes, the door open pin is read to determine if the door is open or not. If it is then a timer is started, this gives you forty five seconds to complete your task before the alarm sounds. With the door closed the timer is stopped and set back to zero. If the sounder goes off while you are rummaging in the fridge the reset button can be pressed, this restarts the timer from zero again.

 

Links and Sources

Extracting MP3 audio from video files

Here is a small Bash script that converts any supported ffmpeg video format; such as .MKV, .MP4 or .MOV and extracts the audio to an .MP3 file, It will also split that MP3 file into chunks and put them in a convenient directory. You will need to install ffmpeg and mp3splt for your particular platform.

Example Usage:

This uses ffmpeg to convert “big fat file.mkv” to “big fat file.mp3” and then uses mp3splt to create a directory “big fat file” containing the files 01 – big fat file.mp3, 02 – big fat file.mp3, etc. The MP3 files will be encoded at 128k Constant Bit Rate and each file will be around 50 minutes in length. To install in Debian/Ubuntu use: sudo apt-get install ffmpeg mp3splt

mp3splt can find the audio in a quiet region near where the split is desired rather than midway through a word, this should make for much cleaner playback across tracks.

Alternative Method

This script gives the same results but uses ffmpeg to split the large MP3 file and then adds track numbering metadata using id3v2. To install in Debian/Ubuntu use: sudo apt-get install ffmpeg id3v2

Creating an Audiobook

Taking this further, I was thinking that it would be nice to have these converted into the M4B Audiobook format for use on my elderly iPod. The script below assumes that you have processed the files as above and have added metadata tags using a tool like mp3tag (yes I know this is for Windows).

To complete this we need to: Combine the multiple MP3 files into one big file, or read the original big file then convert that to M4B format at 96K bit and add chapter marks every ten minutes. For this I have used ffmpeg v3.2.12 and libmp4v2 (for the mp4chaps utility), to install in Debian/Ubuntu use: sudo apt-get install libmp4v2-dev mp4v2-utils ffmpeg

This script works best from a single MP3 file rather than from those that have been re-combined back into a single file, recombining the files caused ffmpeg to exclaim “invalid packet size” and “invalid data” errors. It is able to tell the difference between a directory and a single MP3 and processes the file accordingly, don’t forget to add metadata tags and cover art before you run the script.

When encoding to the M4B using a re-combined file I saw a few of these errors from ffmpeg:

These appear to be caused by the mp3splt program from when the original MP3 file was being split into 50 minute chunks, but I can’t hear any effect on the output.

Lots of information about the file can be gotten using mediainfo, to install in Debian/Ubuntu use: sudo apt-get install mediainfo, example use:

Links and References