105 lines
3.8 KiB
Python
105 lines
3.8 KiB
Python
import cv2
|
|
import numpy as np
|
|
import matplotlib.pyplot as plt
|
|
|
|
|
|
class VideoProcess:
|
|
def __init__(self, path):
|
|
# Used for running average calculation
|
|
self.alpha = 0.9
|
|
# Unused for now. Working on ideas for other potential processing algos
|
|
self.bucketSize = 2
|
|
# Unused for now. Working on ideas for other potential processing algos
|
|
self.frames = []
|
|
# Frame counter
|
|
self.frame = 0
|
|
# Opencv Capture object
|
|
self.cap = cv2.VideoCapture(path)
|
|
# Amount to reduce the video res by
|
|
self.scalingFactor = 4
|
|
# Rolling average of the frame to smooth out comparison and random passers by
|
|
self.frameAverage = None
|
|
# Array of the counts of diff pixels which represents motion
|
|
self.counts = []
|
|
# Rolling average of the counts
|
|
self.aveCount = 0.0
|
|
# Used to determine rolling average size
|
|
self.aveCountAlpha = 1.0 - 1.0 / 10.0
|
|
# place to store rolling average
|
|
self.aveCounts = []
|
|
# Place to store calculated times given the FPS
|
|
self.times = []
|
|
# the FPS of the camera and thus the signal rate for the sampler
|
|
self.FPS = 30.0
|
|
|
|
# Single step in the process getting a motion count
|
|
def processFrame(self):
|
|
# Making sure frames left in video
|
|
if self.cap.isOpened():
|
|
# Read a frame
|
|
ret, src = self.cap.read()
|
|
# Bails if no frames
|
|
if not ret:
|
|
return False
|
|
|
|
# Get the height and width of the video
|
|
h, w, c = src.shape
|
|
# Scale
|
|
src = cv2.resize(
|
|
src, (int(w / self.scalingFactor), int(h / self.scalingFactor))
|
|
)
|
|
# Convert to gray
|
|
gray = cv2.cvtColor(src, cv2.COLOR_BGR2GRAY)
|
|
# Populate the frame average with the first frame
|
|
if self.frameAverage is None:
|
|
self.frameAverage = gray
|
|
|
|
# Wait till there are at least a few frames to compare
|
|
if len(self.frames) < self.bucketSize:
|
|
self.frames.append(gray)
|
|
else:
|
|
self.frames.append(gray)
|
|
self.frames.pop(0)
|
|
# Rolling average calculation
|
|
self.frameAverage = self.frameAverage * self.alpha + self.frames[0] * (
|
|
1.0 - self.alpha
|
|
)
|
|
self.frameAverage = np.uint8(self.frameAverage)
|
|
# Calculate the absolute difference between a current frame and the rolling average
|
|
# This is the magic
|
|
diff = cv2.absdiff(gray, self.frameAverage)
|
|
# Normalize the count to the maximum of pixel value
|
|
count = np.sum(diff) / (w * h * 255 / self.scalingFactor) * 100.0
|
|
|
|
# Show the result
|
|
cv2.imshow("src", src)
|
|
cv2.imshow("diff", diff)
|
|
# Unused live charting the results
|
|
# if self.frame % 10 ==0:
|
|
# self.drawPlot()
|
|
|
|
# Capture the rolling average, non-average count, and timestamps for analysis.
|
|
self.counts.append(count)
|
|
self.aveCount = self.aveCount * self.aveCountAlpha + count * (
|
|
1.0 - self.aveCountAlpha
|
|
)
|
|
self.aveCounts.append(self.aveCount)
|
|
self.times.append(self.frame / self.FPS)
|
|
|
|
# Loop control
|
|
key = cv2.waitKey(1)
|
|
if key == 27:
|
|
return False
|
|
# Iterate frame count
|
|
self.frame += 1
|
|
else:
|
|
plt.show(block=True)
|
|
return False
|
|
return True
|
|
|
|
def drawPlot(self):
|
|
plt.cla()
|
|
plt.plot(self.times, self.counts)
|
|
plt.plot(self.times, self.aveCounts)
|
|
plt.pause(0.0001)
|