OpenCVで、動き検出 | 宇宙仮面の研究室

【爆速】OpenCVで複数の物体検出　－ラズパイ－を参考に、物体認識を試してみた。

単一オブジェクトの動き検出

import numpy as np 
import cv2

ESC_KEY = 27

def get_bounding_box(X, Y):
    x_max = int(np.max(X))
    x_min = int(np.min(X))
    y_max = int(np.max(Y))
    y_min = int(np.min(Y))
    return (x_min, y_min, x_max, y_max)

def draw_bounding_box(img, bbox):
    img = cv2.rectangle(img, (bbox[0], bbox[1]), (bbox[2], bbox[3],), (0, 255, 0), 3)
    return img

fgbg = cv2.bgsegm.createBackgroundSubtractorMOG()

cap = cv2.VideoCapture(0) 
w = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
h = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
fps = cap.get(cv2.CAP_PROP_FPS)

while True:  
    ret, frame = cap.read()
    if frame is None:
        break
    frame = cv2.GaussianBlur(frame, (7,7), 0)

    fgmask = fgbg.apply(frame)

    img1=frame
    img2=fgmask

    Y, X = np.where(fgmask &gt; 200)
    if len(X) == 0 or len(Y)==0 :
      continue

    bbox = get_bounding_box(X, Y)
    img2 = draw_bounding_box(img1, bbox)

    cv2.imshow('img1', img2)
    cv2.imshow('fgmask', fgmask)
    
    if cv2.waitKey(30) == ESC_KEY:
        break

cap.release()
cv2.destroyAllWindows()

import numpy as np

import cv2

ESC_KEY = 27

def get_bounding_box(X, Y):

x_max = int(np.max(X))

x_min = int(np.min(X))

y_max = int(np.max(Y))

y_min = int(np.min(Y))

return (x_min, y_min, x_max, y_max)

def draw_bounding_box(img, bbox):

img = cv2.rectangle(img, (bbox[0], bbox[1]), (bbox[2], bbox[3],), (0, 255, 0), 3)

return img

fgbg = cv2.bgsegm.createBackgroundSubtractorMOG()

cap = cv2.VideoCapture(0)

w = cap.get(cv2.CAP_PROP_FRAME_WIDTH)

h = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)

fps = cap.get(cv2.CAP_PROP_FPS)

while True:

ret, frame = cap.read()

if frame is None:

break

frame = cv2.GaussianBlur(frame, (7,7), 0)

fgmask = fgbg.apply(frame)

img1=frame

img2=fgmask

Y, X = np.where(fgmask > 200)

if len(X) == 0 or len(Y)==0 :

continue

bbox = get_bounding_box(X, Y)

img2 = draw_bounding_box(img1, bbox)

cv2.imshow('img1', img2)

cv2.imshow('fgmask', fgmask)

if cv2.waitKey(30) == ESC_KEY:

break

cap.release()

cv2.destroyAllWindows()

KMeans を使って複数オブジェクトを認識をしてみたところ、ノイズがあると重心がうまく分離できず、Bounding Box が大きくずれてしまう。また、同一の物体であっても、複数のクラスターから構成されてしまうという問題点があった。

import numpy as np 
import cv2
from sklearn.cluster import KMeans

CLUSTER=10
colorlist = [(0,0,0),(0,0,255),(0,255,0),(0,255,255),(255,0,0),(255,0,255),(255,255,0),(255,255,255)]
ESC_KEY = 27

def get_bounding_box(X, Y, result, cluster):
    NO = np.where(result==cluster)
    x_max = int(np.max(X[NO]))
    x_min = int(np.min(X[NO]))
    y_max = int(np.max(Y[NO]))
    y_min = int(np.min(Y[NO]))
    return (x_min, y_min, x_max, y_max)

def draw_bounding_box(img, bbox):
    img = cv2.rectangle(img, (bbox[0], bbox[1]), (bbox[2], bbox[3],), (0, 255, 0), 3)
    return img

def draw_circle(img, X, Y, result, cluster):
    NO = np.where(result==cluster)
    for p in NO[0] :
      img = cv2.circle(img, (X[p], Y[p]), 5, colorlist[cluster%8],-1)
    return img

font = cv2.FONT_HERSHEY_SIMPLEX
def draw_number(img, X, Y, result, cluster):
    NO = np.where(result==cluster)
    for p in NO[0] :
      img = cv2.putText(img, str(cluster),(X[p], Y[p]), font, 1,colorlist[cluster%8],1,cv2.LINE_AA)
    return img

fgbg = cv2.bgsegm.createBackgroundSubtractorMOG()

cap = cv2.VideoCapture("video/vtest.avi") 
w = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
h = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
fps = cap.get(cv2.CAP_PROP_FPS)

while True:  
    ret, frame = cap.read()
    if frame is None:
        break
    frame = cv2.GaussianBlur(frame, (5,5), 0)

    fgmask = fgbg.apply(frame)

    img1=frame
    img2=fgmask

    Y, X = np.where(fgmask &gt; 200)

    if X.size &lt; w :
      continue

    means = KMeans(n_clusters=CLUSTER, init='k-means++', n_jobs=1).fit_predict(np.array([X,Y]))

    for i in range(CLUSTER) :
      bbox = get_bounding_box(X, Y, means, i)
      
      img2 = draw_bounding_box(img1, bbox)
      img2 = draw_circle(img1, X, Y, means, i)
      <em># img2 = draw_number(img1, X, Y, means, i)</em>
      
    cv2.imshow('img1', img2)
    cv2.imshow('fgmask', fgmask)
    
    keyboard = cv2.waitKey(30)
    if keyboard == ESC_KEY:
        break

cap.release()
cv2.destroyAllWindows()

import numpy as np

import cv2

from sklearn.cluster import KMeans

CLUSTER=10

colorlist = [(0,0,0),(0,0,255),(0,255,0),(0,255,255),(255,0,0),(255,0,255),(255,255,0),(255,255,255)]

ESC_KEY = 27

def get_bounding_box(X, Y, result, cluster):

NO = np.where(result==cluster)

x_max = int(np.max(X[NO]))

x_min = int(np.min(X[NO]))

y_max = int(np.max(Y[NO]))

y_min = int(np.min(Y[NO]))

return (x_min, y_min, x_max, y_max)

def draw_bounding_box(img, bbox):

img = cv2.rectangle(img, (bbox[0], bbox[1]), (bbox[2], bbox[3],), (0, 255, 0), 3)

return img

def draw_circle(img, X, Y, result, cluster):

NO = np.where(result==cluster)

for p in NO[0] :

img = cv2.circle(img, (X[p], Y[p]), 5, colorlist[cluster%8],-1)

return img

font = cv2.FONT_HERSHEY_SIMPLEX

def draw_number(img, X, Y, result, cluster):

NO = np.where(result==cluster)

for p in NO[0] :

img = cv2.putText(img, str(cluster),(X[p], Y[p]), font, 1,colorlist[cluster%8],1,cv2.LINE_AA)

return img

fgbg = cv2.bgsegm.createBackgroundSubtractorMOG()

cap = cv2.VideoCapture("video/vtest.avi")

w = cap.get(cv2.CAP_PROP_FRAME_WIDTH)

h = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)

fps = cap.get(cv2.CAP_PROP_FPS)

while True:

ret, frame = cap.read()

if frame is None:

break

frame = cv2.GaussianBlur(frame, (5,5), 0)

fgmask = fgbg.apply(frame)

img1=frame

img2=fgmask

Y, X = np.where(fgmask > 200)

if X.size < w :

continue

means = KMeans(n_clusters=CLUSTER, init='k-means++', n_jobs=1).fit_predict(np.array([X,Y]))

for i in range(CLUSTER) :

bbox = get_bounding_box(X, Y, means, i)

img2 = draw_bounding_box(img1, bbox)

img2 = draw_circle(img1, X, Y, means, i)

<em># img2 = draw_number(img1, X, Y, means, i)</em>

cv2.imshow('img1', img2)

cv2.imshow('fgmask', fgmask)

keyboard = cv2.waitKey(30)

if keyboard == ESC_KEY:

break

cap.release()

cv2.destroyAllWindows()

次に、オブジェクト輪郭検出を参考に複数のオブジェクトの動き認識してみた。動いていないと認識できない、物体の特定ができない、といった問題があり、正確ではないけれど、ある程度認識することができる。

import numpy as np 
import cv2

ESC_KEY = 27

fgbg = cv2.bgsegm.createBackgroundSubtractorMOG()

cap = cv2.VideoCapture("video/vtest.avi") 
w = cap.get(cv2.CAP_PROP_FRAME_WIDTH)
h = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
fps = cap.get(cv2.CAP_PROP_FPS)

while True:  
  ret, frame = cap.read()
  if frame is None:
      break
  frame = cv2.GaussianBlur(frame, (7,7), 0)

  fgmask = fgbg.apply(frame)

  # find contours
  labels, contours, hierarchy = cv2.findContours(fgmask, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)

  for i in range(0, len(contours)):
      if len(contours[i]) > 0:

          # remove small objects
          if cv2.contourArea(contours[i]) &lt; 500:
              continue

          # Draw Bounding Box
          rect = contours[i]
          x, y, w, h = cv2.boundingRect(rect)
          cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 3)

          # Draw Contours
          # frame = cv2.drawContours(frame, contours, i, (0,255,0), 3)

  cv2.imshow('frame', frame)
  cv2.imshow('fgmask', fgmask)
  
  keyboard = cv2.waitKey(int(1000/fps))
  if keyboard == ESC_KEY:
      break

cap.release()
cv2.destroyAllWindows()

import numpy as np

import cv2

ESC_KEY = 27

fgbg = cv2.bgsegm.createBackgroundSubtractorMOG()

cap = cv2.VideoCapture("video/vtest.avi")

w = cap.get(cv2.CAP_PROP_FRAME_WIDTH)

h = cap.get(cv2.CAP_PROP_FRAME_HEIGHT)

fps = cap.get(cv2.CAP_PROP_FPS)

while True:

ret, frame = cap.read()

if frame is None:

break

frame = cv2.GaussianBlur(frame, (7,7), 0)

fgmask = fgbg.apply(frame)

# find contours

labels, contours, hierarchy = cv2.findContours(fgmask, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)

for i in range(0, len(contours)):

if len(contours[i]) > 0:

# remove small objects

if cv2.contourArea(contours[i]) < 500:

continue

# Draw Bounding Box

rect = contours[i]

x, y, w, h = cv2.boundingRect(rect)

cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 3)

# Draw Contours

# frame = cv2.drawContours(frame, contours, i, (0,255,0), 3)

cv2.imshow('frame', frame)

cv2.imshow('fgmask', fgmask)

keyboard = cv2.waitKey(int(1000/fps))

if keyboard == ESC_KEY:

break

cap.release()

cv2.destroyAllWindows()

drawContours だけで輪郭を描画できる。

コメントを残す コメントをキャンセル

コメントを残すコメントをキャンセル