0

I have some video footage and wish to analyse when a light source blinks. The light source is at the same location, so should easy to work with the ROI.

I'm used to working with python, but not very strong when it comes to video analyzing - don't care that much about formats and technical stuff. Just wish to find a quick a dirty way of detecting this.

My current approach would be something like this

  1. Load the video
  2. Extract an image from the video in order to elect region of interest (ROI) by drawing a rectangle (do hope there is an easy module for this)
  3. Go through the entire video to detect pixel changes in the ROI
  4. Log/plot the time when the changes exceed a treshold

I'm pretty sure someone has done something similar, so links to any useful modules, tutorials or handy software would be great. Thanks.

Norfeldt
  • 8,272
  • 23
  • 96
  • 152

2 Answers2

4
import cv2
import numpy as np
import matplotlib.pyplot as plt

# Load the video
cap = cv2.VideoCapture('../data/video/video_test.mp4')

# Extract an image from the video in order to elect region of interest (ROI) by drawing a rectangle
ret, frame = cap.read()

# Draw a rectangle on the image
cv2.rectangle(frame, (100, 100), (300, 300), (255, 0, 0), 2)

# Select the ROI
roi = frame[100:300, 100:300]

# Convert the image to grayscale
roi_gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)

# Create a mask
roi_mask = np.zeros(roi_gray.shape[:2], np.uint8)

# Draw a circle on the mask
cv2.circle(roi_mask, (150, 150), 100, 255, -1)

# Apply the mask to the grayscale image
masked_roi_gray = cv2.bitwise_and(roi_gray, roi_gray, mask=roi_mask)

# Create a threshold to exclude minute movements
threshold = 70

# Apply threshold to the masked grayscale image
_, thresh = cv2.threshold(masked_roi_gray, threshold, 255, cv2.THRESH_BINARY)

# Find changes in the masked grayscale image
_, contours, _ = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

# Draw a circle around the detected changes
for cnt in contours:
    (x, y, w, h) = cv2.boundingRect(cnt)
    cv2.circle(roi, (x + int(w/2), y + int(h/2)), 5, (0, 0, 255), -1)

# Show the image
cv2.imshow('image', frame)

# Wait for a key press
cv2.waitKey(0)

# Destroy all windows
cv2.destroyAllWindows()
Leoli
  • 719
  • 1
  • 9
  • 18
zorg
  • 41
  • 7
1

mahotas may be a good choice for analyzing the image. It loads an image as a numpy array so it would be trivial to elect a ROI. Also it has built-in methods to threshold, calculate mean brightness of picture and stuff like this. Last but not least, mahotas' documentation is pretty good.

I do not know the best way to extract frames from a video in Python (though you can done it with something like opencv but it seems like an overkill) so I suggest using some external program such as ffmpeg and calling it with subprocess module. Also, quick googling gave me this which may be suitable.

Qumeric
  • 518
  • 5
  • 16