It depends on your use case. Vision tasks (object detection, image classification) need a camera. Audio tasks (keyword spotting, sound classification) need a microphone via I2S or ADC. Motion tasks (gesture recognition, fall detection) use IMU or accelerometer via I2C/SPI. Predictive maintenance typically uses vibration or current sensors.
Can I connect external sensors to any MCU?
Most sensors communicate via standard interfaces: I2C, SPI, or analog ADC. Nearly all edge-AI-capable MCUs support these interfaces. Camera sensors require specific peripheral support (DVP, DCMI, or USB) which is only available on select MCUs like ESP32-S3, STM32H7, or STM32F4.
How does sensor selection affect model requirements?
Different sensors produce different data volumes. Camera input requires the most RAM and processing power — 256 KB+ SRAM and 240+ MHz clock. Audio processing needs 64-128 KB. Accelerometer and temperature data are compact and can run on MCUs with as little as 32-64 KB SRAM.
What if I have sensors not listed here?
The 9 sensor types cover the most common edge AI inputs. Specialized sensors (radar, lidar, pressure) can often be mapped to similar categories. Radar is similar to camera for processing requirements, while pressure sensors behave like temperature sensors in terms of data volume.