Hi all,
that's good to read people wanting to play with AI with our ST ToF sensors.
I hope I will give you more than hints in this answer.
STSW_IMG035 is an all-in-one package for our Gesture solutions. Inside, you will find 3 kinds of "Gestures":
- hand tracking (analytic solution)
- motion gestures (analytic solution)
- hand postures ( AI-based solution)
Let's focus on our Hand posture ToF AI solution.
In the STSW_IMG035, we are showing an example of AI model supporting 7 postures and trained with VL53L8CX data. To be honest, the accuracy is very good, and you can feel the power of AI with ST low resolution ToF data.
How did we create this model?
Everything is available on the STM32AI ModelZOO - GitHub, stm32ai-modelzoo/hand_posture at main · STMicroelectronics/stm32ai-modelzoo · GitHub
STM32AI ModelZOO - GitHub
When I'm saying everything, it really means everything:
- Python scripts (tensorflow / Keras / ...)
- ST AI model topology (2D CNN)
- ST training Dataset (used to generate the AI model of STSW_IMG035)
The documentation is embedded in GitHub (READ.ME) and if you are motivated to read it, I have no doubt you will be able to customize your own AI model.
The ST ModelZOO is sharing a common Python environment between all use cases, you just have to follow the installation procedure described on the welcome page: "Before you start"
Hand posture ToF AI - web interface
We also released a graphical AI tool which is part of the famous ST Edge AI Suite , called Hand posture ToF AI.
This web interface is using directly the STM32AI modelZOO scripts, it's easy to use and you can test your model directly on the web page if your board is connected to your computer. Just follow the steps.
You can't modify everything like STM32AI ModelZOO, but it allows to modify the basic AI parameters.
The documentation is embedded in the web page.
One last information, the STSW_IMG035 is the official data logger to capture your data in case you would like to create your dataset or create new postures (it's explained in the documentations)
I hope this will help you start your journey into the wonderful world of AI. Our ToF sensors can do much more than measuring distances thanks to AI.
Our community relies on fruitful exchanges and good quality content. You can thank and reward helpful and positive contributions by marking them as 'Accept as Solution'. When marking a solution, make sure it answers your original question or issue that you raised.
ST Employees that act as moderators have the right to accept the solution, judging by their expertise. This helps other community members identify useful discussions and refrain from raising the same question. If you notice any false behavior or abuse of the action, do not hesitate to 'Report Inappropriate Content'
