Ixiko / Camera-Motion-Detection-Tests

some test with Python and OpenCV on Windows

Motion-Detection-Tests

Python script(s) to test OpenCV capabilities for motion detection

To try this creepy little script you need the following libraries:

Operating system dependent libraries

• OpenCV 4.6+ from here Releases - OpenCV - choose your platform. This script was written on Windows. I can't remember what the name of the python wrapper libraries was. If you can't remember or never knew, here is the pip command line and the name:

  pip install OpenCV-Python

• 2 other important libraries are numpy and imutils these can be obtained with pip using:

  pip install numpy
  pip install imutils

• I use the library mjpeg_streamer to have a simple web server to stream the output:

  pip install mjpeg_streamer

• Don't forget to update your libraries from time to time by adding the --upgrade attribute to them

  pip install what_I_already_installed_to --upgrade

Operating system dependent libraries

• Windows system

  pip install pywin32
  pip install pygetwindow  

• MacOS system

pip install pyobjc-framework-Quartz 

• Linux system

pip install python-xlib

This is really a lot of code for a small amount of features.

---

How to see the 3 motion detecting algorithms (ways)

• you only need to change this variable to 1, 2 or 3 in main script:

  analyse_mode    = 1          # 1 = contour, 2 = eraseBackground, 3 = mask_motion

---

The most fundamental things I have found and noticed:

• To use the maximum video size of your camera you have to tell OpenCV which backend to use. The VideoCaptureApi can be set on windows to one of these:

  • CAP_DSHOW = 700

  • CAP_MSMF = 1400

  cv2.CaptureVideo(0, cv2.CAP_DSHOW)

• and this is how you set the width and height of the video

   cap.set(cv2.CAP_PROP_FRAME_WIDTH , cam_width)
   cap.set(cv2.CAP_PROP_FRAME_HEIGHT, cam_height)

• A captured image should be converted to a gray image as it is more effective for calculations than leaving it in colour.

  grayImg = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

• Blurring images allows for better contour detection (is that so?)

  grayImg = cv2.GaussianBlur(grayImg, (21, 21), 0)

• If I increase the exposure of my webcam, the frame rate I get from the camera decreases noticeably.

---

These things needs a solution:

• The image noise of the camera is often recognised as movement. Which is the most effective way to reduce noise? Is it ultimately more effective to take better hardware?

• Motion detection becomes unreliable when the camera is in automatic mode for brightness, contrast, saturation or exposure. **Solution **found on Windows: by using cv2.CAP_DSHOW and set the camera property cap.set(cv2.CAP_PROP_SETTINGS, 0) , windows will open a dialog window.

---

Question(s) left:

• when using the erase Background feature of OpenCV, a 'ghost image' is always calculated, Even if nothing has changed in the scene.