← Back
Stable Advanced Edge AI

Vision Node (TinyML)

On-device person detection in 54 ms — the camera thinks before it phones home.

Board
Seeed XIAO ESP32-S3 Sense
Est. cost
~$28
Install time
60 min
Sensors
OV2640 camera, PDM microphone, TFLite Micro (person detection)

EspCamML puts TensorFlow Lite Micro on the XIAO ESP32-S3 Sense: the OV2640 camera and PDM microphone feed on-device models, with ESP-NN SIMD acceleration bringing person detection down to ~54 ms. The agent gets tools like camera_classify ("person 0.92"), sound_detect, and background monitors that watch without any LLM in the loop. Cloud escalation (describe an image, transcribe audio, send to Telegram) is optional — the default is local, private, fast.

Maintained and field-tested by the Latent team.

Start guided install

What you need

Seeed XIAO ESP32-S3 Sense (8 MB PSRAM + camera + mic) ~$22
USB-C cable ~$3
MicroSD card (optional, for clips) ~$3
Est. cost~$28

Install steps

1
Start from the agent runtime
2
Build with the camera profile
3
Flash and run the smoke tests
4
Set up a background monitor

Device config

Full config — copy it, download it, or follow the guided install.

# EspCamML — Vision Node with on-device person detection (TinyML) # Source: ~/MyBOT/EspCamML · Board: Seeed XIAO ESP32-S3 Sense # Builds on the zclaw agent runtime + esp-tflite-micro (ESP-NN accelerated). # # 1. Build with the camera profile on top of zclaw cd ~/MyBOT/zclaw idf.py -D SDKCONFIG_DEFAULTS="sdkconfig.defaults;sdkconfig.esp32s3-cam.defaults" \ set-target esp32s3 # 2. Build + flash + monitor idf.py build idf.py -p /dev/cu.usbmodem1101 flash monitor # On-device tools (all local, no cloud round-trip): # camera_capture, camera_motion_check < 100 ms # camera_classify (person) ~ 54 ms with ESP-NN SIMD # sound_level, sound_detect, sound_tone < 100 ms # monitor_person / monitor_sound background watchers # Cloud escalation (optional): camera_describe, mic_transcribe, # camera_send_telegram. # # Pin map is preconfigured for the XIAO ESP32-S3 Sense OV2640 + PDM mic.

Reviews (0)

No reviews yet — be the first.

Sign in to use this feature — it takes 20 seconds and it’s free. Sign in