1. Repo Structure
2. Useful 3rd Party Apps for This Repo
   1. FFMPEG
3. Description of the Code
   1. switch_colored_bw()
   2. gaussian_and_bilateral()
   3. grab_them_fish()
   4. in_between_scenes()
   5. sobel_filter()
   6. detect_circles()
   7. track_object()

• Example videos are located inside videos folder
• out folder is used by main.py to store the images
• kids.mp4 is a video from shutterstock, in which someone takes a piece of salmon with their fork from their plate

FFMPEG is a great tool for handling videos and images to create your own videos. Installing it on Linux (like Ubuntu) is quite easy, but installing on Windows has some extra steps. To install it on Windows, check out wikihow page for installing FFmpeg on Windows.

FFMPEG will be very useful to create videos from images and also to stitch videos to combine them into one video. In main.py line 155:

filename = f"out/IMG{i:04d}.jpg"
cv2.imwrite(filename=filename, img=frame)

When we set the name of the images this way (which will generate images with ordered names, e.g., IMG0001.jpg), we can use FFMPEG to generate a video with frame rate of 25 from the images inside the current folder with the following line on bash (Linux):

ffmpeg -i img%03d.png -r 25 out.mp4 # Example

For Windows, it should be:

ffmpeg -i img%03d.png -r 25 "../videos/out.mp4"

The code consist of 7 different methods:

• swith_colored_bw()
• manipulate_video()
• grab_them_fish()
• in_between_scenes()
• sobel_filter()
• detect_circles()
• track_object()

After line 603 (if __name__ == '__main__':), these methods are placed commented. You can run them all or one by commenting/uncommenting them.

The video switches back and forth random from colored to grayscale. The reason for using the cw2.imwrite() method instead of using the built-in cv2.VideoWriter class is that the cv2.VideoWriter class has a parameter colored which can be set to True or False, meaning that you can only save colored or grayscale videos.

• Possible work around that I haven't tried: try passing the grayscale image as 3-channel image by np.stack((grayscale_image,)*3, axis=-1) to see if cv2.VideoWriter class can save both colored and grayscale images.

• I used time library's time.time() method for handling the time, but using cv2.imwrite() gives the perk of setting exactly how many images you want to write which in return you can decide the duration of the video by setting the framerate with FFMPEG. Trying to use cv2.VideoWriter class with time.time() method has the issue of not setting the timing correctly, albeit using cv2.imwrite() having the issue of missing frames.

This method edits the video kids.mp4 from ShutterStock, (link: https://www.shutterstock.com/nb/video/clip-11921927-pick-grilled-salmon-dish-olives-baguette ) to show the difference between bilateral filter and Gaussian filter.

This method tries to use two color filters (RGB and HSV) to mask images and compare the performance of those filters as well the performance when "Closing" morphology operation is applied to masks. You will see that HSV-based filter will perform better (will filter more of the image) than the RGB one. This could be explained due to the fact that "Hue" is a much better filter for colors than individual RGB values.

This method creates a black background with text to divide your episodes (if you ever need it)

This method captures your webcam stream and applies sobel_x, sobel_y, and sobel_xy to video stream.

This method uses OpenCV library's built-in HoughCircles() method to obtain the detected circles' radii and centers. Futhermore, this method captures your webcam stream.

This method tries to track the object via a "primitive", counting-based, probabilistic method. It also blends your webcam stream and the tracking probabilty.