I wanted a home camera without a vendor's cloud, app or subscription, so I wrote my own firmware: live view in the browser, motion recording to microSD (with a few seconds before the motion), face recognition on the chip with a "who was seen" log, a recordings browser with MP4 download, and optional remote access through a small relay on my own VPS.
WHY THE ESP32-S3
I started on the classic AI-Thinker ESP32-CAM. Streaming and recording worked, but I moved to an ESP32-S3-CAM N16R8 (HW-679) for two reasons:
• Face recognition: Espressif's esp-dl models use the S3's vector instructions. On the classic ESP32, recognition took seconds per frame.
• Memory: 16 MB flash fits two OTA slots plus a partition for remembered faces. 8 MB PSRAM holds frame buffers, a 1.5 MB pre-motion buffer and the model inputs.
I also swapped in an OV3660 sensor (up to 2048×1536).
LIBRARIES, AND WHY
• Arduino core 3.3.12 (PlatformIO): it ships prebuilt esp-dl face detection and recognition models.…