Ultra-Efficient Custom Software for STM32 Processors
BDTI has extensive experience with the STM32 family of processors going back to their introduction in 2006. Members of our team have:
- Designed and developed embedded systems based on the STM32
- Implemented demanding real-time algorithms for vision, audio, and accelerometer data
Recently, a major consumer products company sought our help using the STM32F429 in a low-cost household appliance. BDTI designed and implemented an ultra-efficient embedded vision pipeline that enabled the appliance to identify consumable items selected by the user and optimize the appliance’s operation accordingly. The vision pipeline included image processing, classical computer vision algorithms, and a deep neural network. To enable real-time response utilizing only the Cortex-M4 processor, BDTI optimized the vision pipeline at multiple levels, including developing novel algorithm blocks, writing optimized algorithm kernel code, and exploiting quantization techniques.
In addition, BDTI maintains the TensorFlow Lite for Microcontrollers (TFLM) repository, often used for on-device machine learning on STM32 processors.
If you have a challenging vision or embedded AI application that you’re trying to implement on the STM32 family, please contact us!