iWave’s “iW-RainboW-G34D” is a Pico-ITX dev kit with 5.5-inch display and a CSI cam that runs Linux on an i.MX8M Mini via its “iW-RainboW-G34M-SM” module. The kit supports NXP’s eIQ ML software for image recognition.

iWave Systems has launched a Pico-ITX form-factor development kit based around an unnamed NXP i.MX8M Mini based module that appears to be its SODIMM-style iW-RainboW-G34M-SM. iWave announced the new kit under the name i.MX8M Mini Board and linked to a product page headlined iW-RainboW-G34D.

The kit was announced in a blog post about an iWave demo that uses the kit for a facial recognition system for building entry. The demo relies on NXP’s eIQ machine learning software, which runs directly on i.MX8 processors without using an NPU (see farther below).

iW-RainboW-G34D i.MX8M Mini Board

The layout of the iW-RainboW-G34M-SM module found on the iW-RainboW-G34D kit differs from the iW-RainboW-G34M-SM prototype that was announced last June. Whereas the module was announced with support for both the i.MX8M Mini and i.MX8M Nano, the new version supports only the Mini. In addition, the current iW-RainboW-G34M-SM — and the iW-RainboW-G34D — ship with 1GB to 4GB LPDDR4 compared to 2GB to 8GB RAM in last year’s announcement.


iW-RainboW-G34D i.MX8M Mini Board dev kit
(click image to enlarge)

The iW-RainboW-G34D i.MX8M Mini Board, which follows Estone Technology’s Mini-based, Pico-ITX form factor EMB-2237-AI, runs Linux 4.14.98 or Android Pie 9.0.0 on the iW-RainboW-G34M-SM module. The module is equipped with 8GB or more eMMC, and its 2MB QSPI Flash and microSD slot are optional on the dev kit. The module provides the kit with dual-band 802.11a/b/g/n/ac with BLE v5.0.


iW-RainboW-G34D block diagram (left) and updated image for iW-RainboW-G34M-SM module
(click images to enlarge)

The 100 x72mm carrier board adds 1x or 2x GbE, 2x USB 2.0 host, a micro-USB debug port, and a micro-USB port that is variably described as device or OTG. Other features include an audio I/O jack, JTAG, and GPIO, UART, and eCSPI expansion headers.

The board appears to have an HDMI port, which is not listed in the specs. You also get MIPI-DSI and -CSI connectors, both of which connect to a daughterboard with a 5.5-inch HD AMOLED display and an undocumented MIPI-CSI camera.

The iW-RainboW-G34D kit has a 5V @1A DC jack along with a PMIC and power and reset buttons. Other features include a boot mode switch and an RTC with coin cell. The board has a 0 to 60°C operating range.

 
iWave facial recognition demo

The iWave blog pitches facial recognition as a preferred security solution in the age of Covid-19 compared to contact-based security screening systems, which presumably include card readers, keypads, fingerprint readers, and touchscreen systems. Other tech solutions for avoiding contact infections include Techbase’s planned, Raspberry Pi based Smart Delivery Box, the first in a series of open source #CoronaIoT projects.

The blog post does not discuss the system’s accuracy when people are wearing masks, which would likely be the case in a situation where one is concerned about contact-based infections. Last month, a Security Magazine story suggested that facial recognition vendor claims of being able to recognize faces when wearing masks should be greeted with skepticism. High-end AI systems can achieve relatively high occlusion detection accuracy, assuming they already possess an image of the individual wearing a mask, but most solutions fare much worse.

iWave’s facial recognition demo workflow (left) and NXP’s eIQ inference support diagram
(click images to enlarge)

The iWave demo does not include a separate NPU. The facial recognition algorithm runs directly on the i.MX8M Mini using NXP’s eIQ OpenCV ML software. eIQ runs on all of NXP’s i.MX8 SoCs, as well as its MCU-like, uClinux-ready i.MX RT crossover SoCs.

eIQ supports a variety of inference engines and offers neural network compilers and optimized libraries. When running on i.MX8 family processors, eIQ defaults to the open source, Linux-based Arm NN SDK and inference engine with a featured option for TensorFlow Lite. It uses Arm CMSIS-NN on the RT processors. The eIQ software “leverages open-source technologies and is fully integrated into our MCUXpresso SDK and Yocto development environments,” says NXP.

On selected configurations of the i.MX8X and i.MX8 Quad, NXP’s eIQ could perhaps tap into the 640MHz HiFi4 DSP for AI acceleration. It also supports the upcoming i.MX8M Plus, which offers the DSP, dual ISPs, and a 2.3 TOPS NPU.

The i.MX8M Mini lacks any of these accelerators, so eIQ instead uses the SoC’s Arm NEON extension for acceleration. You get 1x, 2x, or 4x Cortex-A53 (1.8GHz) and single Cortex-M4 (400MHz) cores. The Mini also provides GCNanoUltra (3D) and GC320 (2D) graphics cores with HD video support.

In the iWave demo, images are captured via MIPI-CSI, and the eIQ software compares the novel face features with the known face images in the Django framework database. Real-time data logging can be configured to send the information along with an alert.

 
Further information

No pricing or availability information was provided for the iW-RainboW-G34D i.MX8M Mini Board. More information may be found in iWave Systems’ blog announcement featuring the facial recognition demo, as well as the iW-RainboW-G34D product page.