AI Vision System

[insert placeholder image when done]

The Premise

I'll be designing a compact product, that'll be able to describe a scene presented to the camera, giving a description to aid a person who is visually impaired.

Criteria for Success

Safe to Use
- Doesn't get hot in the users hands
- Isn't sharp
- Is able to be easily carried around
- Is durable, and able to take some damage without posing a safety risk due to sharp edges or burs that may appear.
Convenient
- The user actually wants to use it
  - Fast load times (under 5 seconds from button press to an accurate and detailed description)
- Long lasting power solution
Accesible
- Easy to press button
- Physical feedback to the user
- Doesn't require cellular connectivity.

Day 1 - Definition

So.. to start off, the main controller. My reaserch indicates that I essentially have 3 main options: - Raspberry Pi CM5 - Orange Pi CM5 - Radxa CM5

I'm ultimately very fond of the compute module as a form factor, while packing in an insane amount of power.

After comparing the 3, the RPi version was quickly eliminated. It simply wouldn't be able to provide the required processing time, making the user experience slow and buggy.

When choosing between the Radxa and Orange Pi, I ultimately did some research on the community available support, as well as availability. This yielded the Radxa to be the most suitable solution. Based on how things go with the budget, at the end, I'll either choose between the 8GB or 16GB RAM versions. This gives a bit of wiggle room in terms of design.

Currently, my main focus will be on the PCB and Case, working on the software once that is done.

The case will be made from 2 pieces of billet aluminium, serving as a heatsink, to allow for optimal operation. More on this to come.

Radxa CM5

Radxa CM5 module

Time Spent 3 Hours (10/07/25)

Day 2 - Starting on the Schematic

Today, I started off with the schematic. I'll be using just a regular USB-C port to provide both power, and allow for data interaction with the finished product.

As is required, there will be a speaker which is able to convert the text to an audio output. I was largely able to base this schematic off the Adafruit example breakout board. Yet, I'm still investigating how the I2S works with the Radxa module.

And the camera... this took a while. It took ages to find the correct cable, and connector to use, and I eventually settled on this. Radxa's documentation wasn't too clear, so I've reached out to them via email, and hopefully I'll hear back from them tomorrow.

In the mean time, I created the motor driver circuit, using a simple MOSFET to drive the small vibration motor. This was luckily quite a bit.. simpler.

I'm pleased with how things are going so far, and I've set the goal to finish this within the next 5 days, to ensure everything can be built in time for the end of Highway!