Calculating frames per second

In the Reading camera frame and video files section, we saw how we can get some properties from the cv2.VideoCapture object. fps is an important metric in computer vision projects. This metric indicates how many frames are processed per second. It is safe to say that a higher number of fps is better. However, the number of frames your algorithm should process every second will depend on the specific problem you have to solve. For example, if your algorithm should track and detect people walking down the street, 15 fps is probably enough. But if your goal is to detect and track cars going fast on a highway, 20-25 fps are probably necessary. 

Therefore, it is important to know how to calculate the fps metric in your computer vision projects. In the following example, read_camera_fps.py, we are going to modify read_camera.py to output the number of fps. The key points are shown in the following code:

# Read until the video is completed, or 'q' is pressed
while capture.isOpened():
    # Capture frame-by-frame from the camera
    ret, frame = capture.read()

    if ret is True:
        # Calculate time before processing the frame:
        processing_start = time.time()

        # All the processing should be included here
        # ...
        # ...
        # End of processing

        # Calculate time after processing the frame
        processing_end = time.time()

        # Calculate the difference
        processing_time_frame = processing_end - processing_start

        # FPS = 1 / time_per_frame
        # Show the number of frames per second
        print("fps: {}".format(1.0 / processing_time_frame))

    # Break the loop
    else:
        break
 

First, we take the time before the processing is done:

processing_start = time.time()

Then, we take the time after all the processing is done:

processing_end = time.time()

Following that, we calculate the difference:

processing_time_frame = processing_end - processing_start

Finally, we calculate and print the number of fps:

print("fps: {}".format(1.0 / processing_time_frame))