I wanted to get a text when someone comes to my front door. I know you can buy a commercial camera that does this, but I wanted more control over the images.  My garage has a window that faces the front door of my house, so I didn't have to build a weatherproof enclosure.

Overview

The system has a smart cam that detects people and notifies the Raspberry Pi when one is the the frame. A python program on the Raspberry Pi writes the Jpeg to a file and uploads it to Google Drive.  The program then uses Twilio to send a text with a link to the photo to my cell phone. 

Hardware

I am using a Jevois smart camera and a Raspberry Pi. The Jevois camera is an open source smart cam that runs Linux and has a number of built in modules and costs about $50.  I've previously tried to do image recognition using just a webcam on a Raspberry Pi, but the performance wasn't good.

The Jevois comes in a small 3D printed case with a mini-USB connector on the back.  I used a piece of scrap MDF to make a base. I initially wanted to use the Jevois on a tripod so I cut a hole and glued a 1/4-20 nut in the MDF. I screwed the Jevois to the MDF with the suggested screws. The holes in the case aren't tapped so it's a little tricky.

To make a stable base, I took the small cardboard box that the Jevois was shipped in, cut a small hole in the top and glued a 1/4-20 bolt in the top. When the glue was dry, I filled the box with sand and screwed the Jevois on to the box.

For the computer, I used a Raspberry Pi 3B that I had from a previous project and I connected the Jevois via USB. One thing to note, The Raspberry Pi doesn't provide enough power to reliably power the Jevois via the USB port. Jevois sells a pigtail USB that has two standard USB connections - one for data, and one for power - connected to a mini-USB. I used one of these with the data connection going to the PI and the power going to a USB power adapter/wall charger.

Software

The Jevois cam has YOLO as one of the bundled machine vision modules. I've had success with it previously, so I just used the stock version.  

On the Raspberry Pi, the following code controls the camera, saves the image when a person is detected, copies it to Google Drive and texts me.

#!/usr/bin/python3
TWaccount_sid = 'ACXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX'
TWauth_token = 'lotsarandomchars'

import os,sys
import time
import shutil
import cv2
import numpy
import serial
import logging
from datetime import datetime
from twilio.rest import Client
import uploader

def SendParm(cmd):
   ser.write (cmd)
   time.sleep(1)
   line = ser.readline()
   if Headless:
      logging.info(cmd.decode('utf8'))
      logging.info(line.decode('utf8'))
   else:
      print (cmd.decode('utf8'))
      print (line.decode('utf8'))
      
#main -------
Headless=True
#default to not headless when debugging in Idle
if 'idlelib.run' in sys.modules:
   #idle
   Headless=False

thresh=50 # default detection threshold
if len(sys.argv)>1:
   if sys.argv[1]=='-show':
      Headless=False
   elif 'thresh' in sys.argv[1]: 
      thresh_arg=sys.argv[1].split("=")
      thresh=int(thresh_arg[1])

if len(sys.argv)>2:
   if 'thresh' in sys.argv[2]:
      thresh_arg=sys.argv[2].split("=")
      thresh=int(thresh_arg[1])

if os.name == 'nt':
   folder = 'C:\\users\\peter/jevois_capture'
   logfile = 'c:\\users\\peter/jevois.log'
   cfg_path = 'c:/Users/peter/frontdoorcam/uploader.cfg'
   port = 'COM7'
   camno = 0
else: #linux
   folder = '/home/pi/jevois_capture'
   logfile = '/home/pi/jevois.log'
   cfg_path = '/home/pi/frontdoorcam/uploader.cfg'
   port = '/dev/ttyACM0'
   camno = 0

logging.basicConfig(filename=logfile,level=logging.DEBUG)
logging.info('------- Jevois Cam Startup --------')

ser = serial.Serial(port,115200,timeout=0.01)

# No windows in headless
if not Headless:
   cv2.namedWindow("jevois", cv2.WINDOW_NORMAL)
   cv2.resizeWindow("jevois",1280,480) 


camera = cv2.VideoCapture(camno)
...