import numpy as np
import cv2
import sys
from random import randint
# Criar um vetor para gerar cores aleatoriamente (R,G,B)
TEXT_COLOR = (randint(0, 255), randint(0, 255), randint(0, 255))
BORDER_COLOR = (randint(0, 255), randint(0, 255), randint(0, 255))
FONT = cv2.FONT_HERSHEY_SIMPLEX
VIDEO_SOURCE = "videos/walking.avi"
# vetor de background subtractors
BGS_TYPES = ["GMG", "MOG", "MOG2", "KNN", "CNT"]
# definir qual BGS usar, inserindo o número que corresponde ao BGS selecionado no vetor de background subtractors
# 0 = GMG, 1 = MOG, 2 = MOG2, 3 = KNN, 4 = CNT
BGS_TYPE = BGS_TYPES[2]
# Kernel: Elemento estruturante
def getKernel(KERNEL_TYPE):
if KERNEL_TYPE == "dilation":
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
if KERNEL_TYPE == "opening":
kernel = np.ones((3, 3), np.uint8)
if KERNEL_TYPE == "closing":
kernel = np.ones((3, 3), np.uint8)
return kernel
# Filtros morfológicos para redução do ruído
def getFilter(img, filter):
'''
Esses filtros são escolhidos a dedo, apenas com base em testes visuais
'''
if filter == 'closing':
return cv2.morphologyEx(img, cv2.MORPH_CLOSE, getKernel("closing"), iterations=2)
if filter == 'opening':
return cv2.morphologyEx(img, cv2.MORPH_OPEN, getKernel("opening"), iterations=2)
if filter == 'dilation':
return cv2.dilate(img, getKernel("dilation"), iterations=2)
if filter == 'combine':
closing = cv2.morphologyEx(img, cv2.MORPH_CLOSE, getKernel("closing"), iterations=2)
opening = cv2.morphologyEx(closing, cv2.MORPH_OPEN, getKernel("opening"), iterations=2)
dilation = cv2.dilate(opening, getKernel("dilation"), iterations=2)
return dilation
def getBGSubtractor(BGS_TYPE):
if BGS_TYPE == "GMG":
return cv2.bgsegm.createBackgroundSubtractorGMG(initializationFrames=120, decisionThreshold=.8)
if BGS_TYPE == "MOG":
return cv2.bgsegm.createBackgroundSubtractorMOG(history=200, nmixtures=5, backgroundRatio=.7, noiseSigma=0)
if BGS_TYPE == "MOG2":
return cv2.createBackgroundSubtractorMOG2(history=500, detectShadows=True, varThreshold=100)
if BGS_TYPE == "KNN":
return cv2.createBackgroundSubtractorKNN(history=500, dist2Threshold=400, detectShadows=True)
if BGS_TYPE == "CNT":
return cv2.bgsegm.createBackgroundSubtractorCNT(minPixelStability=15, useHistory=True, maxPixelStability=15*60, isParallel=True)
print("Unknown createBackgroundSubtractor type")
sys.exit(1)
# Carregar o video
cap = cv2.VideoCapture(VIDEO_SOURCE)
bg_subtractor = getBGSubtractor(BGS_TYPE)
def SubtractorTracker():
# Inicialização do TRACKER
parameters_shitomasi = dict(maxCorners = 100,
qualityLevel = 0.3,
minDistance = 7)
parameters_lucas_kanade = dict(winSize = (15, 15),
maxLevel = 2,
criteria = (cv2.TERM_CRITERIA_EPS | cv2. TERM_CRITERIA_COUNT, 10, 0.03))
colors = np.random.randint(0,255, (100, 3))
ret, frame = cap.read()
frame_gray_init = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
edges = cv2.goodFeaturesToTrack(frame_gray_init, mask = None, **parameters_shitomasi)
mask = np.zeros_like(frame)
while (cap.isOpened):
ok, frame = cap.read()
if not ok:
print("Frame capture failed, stopping...")
break
bg_mask = bg_subtractor.apply(frame)
fg_mask = getFilter(bg_mask, 'combine')
# Resultado da subtração de fundo
result = cv2.bitwise_and(frame, frame, mask=fg_mask)
# Passamos como parâmetro ao Tracker o resultado do BGS
frame_gray = cv2.cvtColor(result, cv2.COLOR_BGR2GRAY)
new_edges, status, errors = cv2.calcOpticalFlowPyrLK(frame_gray_init, frame_gray, edges, None, **parameters_lucas_kanade)
news = new_edges[status == 1]
olds = edges[status == 1]
for i, (new, old) in enumerate(zip(news, olds)):
a, b = new.ravel()
c, d = old.ravel()
mask = cv2.line(mask, (a,b), (c,d), colors[i].tolist(), 2)
frame = cv2.circle(result, (a,b), 5, colors[i].tolist(), -1)
img = cv2.add(result, mask)
cv2.imshow('BGS + Optical flow', img)
if cv2.waitKey(1) == 13:
break
frame_gray_init = frame_gray.copy()
edges = news.reshape(-1,1,2)
SubtractorTracker()
