Few days ago Google Announces New Update to Mendel Linux Distribution, designed for use on Coral boards such as Dev Board and SoM. The Mendel Linux distribution builds on the Debian foundation and it is fully compatible with this project's repositories (unmodified binary packages and updates from the main Debian repositories are used).
Changes are reduced to perform the compilation and creation of downloaded image of eMMC cards and include components for support Coral platform hardware components. Coral specific components are licensed under Apache 2.0.
Dev Board is a platform for rapid prototyping of hardware systemse based on Google Edge's Tensor Processing Unit (TPU) to accelerate machine learning and neural network operations. SoM (System-on-Module) is one of the turnkey solutions for running applications related to machine learning.
Mendel Linux 4.0 Main New Features
This new version It is the first on the system to be upgraded to Debian 10, bringing the compilation is optimized for embedded systems and contains Debian 10 innovations related to SecureBoot and AppArmor support.
Among the innovations are the support for OpenCV and OpenCL, the use of Device Tree overlays, as well as updating some system components, among which stand out GStreamer, Python 3.7, Linux kernel 4.14, and U-Boot 2017.03.3 bootloader.
Of the specific innovations, the possibility of using the Coral GPU is mentioned (Vivante GC7000) installed on the board to speed up the conversion of pixel data from YUV color model to RGB with a productivity of up to 130 frames per second for video with a resolution of 1080p, which can be useful when using plates to process video from cameras to generate a sequence in YUV format.
To use machine learning to process streaming video and sound on the go, It is proposed to use the multiplatform and open source framework «MediaPipe». The machine learning solutions that Google has for MediaPipe include hand tracking and gesture recognition, multi-hand tracking, face detection, hair segmentation, and object detection.
For example, based on this it is possible to implement a system to recognize and track objects or faces in a video transmitted from a surveillance camera.
Typical ready-made and ready-made machine learning models compiled for Edge TPU processors used in Coral panels continue to be delivered on the project website, but are gradually transferred to the general catalog of publicly available TensorFlow Hub models.
To simplify the development of these solutions based on the Coral and Mendel Linux boards, a guide has been prepared showing how to assemble a smart sorter from available materials that distributes colored and white balls into different baskets using a Raspberry Pi and Coral.
Si you want to know more about it about this release, you can check the details in the original post In the following link.
Download Mendel Linux 4.0
In order to obtain the image of this new version of the system, can be done by going to the official website of the project. The link is this.
Of the packages that are included, Mendel system image provided for Coral Dev Board which also includes a recovery.img, which can record on an SD card and recover a card that does not start.
Google recommends that all users by Coral SBC upgrade your Dev Board or SoM to Mendel Linux 4.0 "Day" ASAP, following the internal documentation on how to display a new system image, which is currently only possible on Linux and macOS platforms.
However, before viewing a new image, users should be aware that all local and system data will be removed during the installation process.