import cv2 import numpy as np import os.path as path import argparse parser = argparse.ArgumentParser() parser.add_argument("-a", "--algorithm", help = "m (or nothing) for meanShift and c for camshift") args = vars(parser.parse_args()) def center(points): x = (points[0][0] + points[1][0] + points[2][0] + points[3][0]) / 4 y = (points[0][1] + points[1][1] + points[2][1] + points[3][1]) / 4 return np.array([np.float32(x), np.float32(y)], np.float32) font = cv2.FONT_HERSHEY_PLAIN class Pedestrain(): def __init__(self, id, frame, track_window): self.id = int(id) x, y, w, h = track_window self.track_window = track_window self.roi = cv2.cvtColor(frame[y:y+h, x:x+w], cv2.COLOR_BGR2HSV) roi_hist = cv2.calcHist([self.roi], [0], None, [16], [0, 180]) self.roi_hist = cv2.normalize(roi_hist, roi_hist, 0, 255, cv2.NORM_MINMAX) self.kalman = cv2.KalmanFilter(4, 2) self.kalman.measurementMatrix = np.array([[1, 0, 0, 0], [0, 1, 0, 0]], np.float32) self.kalman.transitionMatrix = np.array([[1, 0, 1, 0], [0, 1, 0, 1], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32) self.kalman.processNoiseCov = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32) * 0.03 self.measurement = np.array((2, 1), np.float32) self.prediction = np.zeros((2, 1), np.float32) self.term_crit = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1) self.center = None self.update(frame) def __del_(self): print "Pedestrain %d destoryed" % self.id def update(self, frame): hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) back_project = cv2.calcBackProject([hsv], [0], self.roi_hist, [0, 180], 1) if args.get("algorithm") == "c": ret, self.track_window = cv2.CamShift(back_project, self.track_window, self.term_crit) pts = cv2.boxPoints(ret) pts = np.int0(pts) self.center = center(pts) cv2.polylines(frame, [pts], True, 255, 1) if not args.get("algorithm") or args.get("algorithm") == "m": ret, self.track_window = cv2.meanShift(back_project, self.track_window, self.term_crit) x, y, w, h = self.track_window self.center = center([[x, y], [x+w, y], [x, y+h], [x+w, y+h]]) cv2.rectangle(frame, (x, y), (x+w, y+h), (255, 255, 0), 2) self.kalman.correct(self.center) prediction = self.kalman.predict() cv2.circle(frame, (int(prediction[0]), int(prediction[1])), 4, (255, 0, 0), -1) cv2.putText(frame, "ID:%d -> %s" % (self.id, self.center), (11, (self.id +1)*25+1), font, 0.6, (0, 0, 0), 1, cv2.LINE_AA) cv2.putText(frame, "ID:%d -> %s" % (self.id, self.center), (10, (self.id +1)*25), font, 0.6, (0, 255, 0), 1, cv2.LINE_AA) def main(): camera = cv2.VideoCapture(path.join(path.dirname(__file__), "/home/utryjc/Pictures/vtest.avi")) history = 20 bs = cv2.createBackgroundSubtractorKNN() cv2.namedWindow("surveillance") pedestrains = {} firstFrame = True frames = 0 fourcc = cv2.VideoWriter_fourcc(*'XVID') out = cv2.VideoWriter('output.avi', fourcc, 20.0, (640, 480)) while True: print "----------------------FRAME %d------------------" % frames grabbed, frame = camera.read() if (grabbed is False): print "fail to grab frame" break fgmask = bs.apply(frame) if frames < history: frames += 1 continue th = cv2.threshold(fgmask.copy(), 127, 255, cv2.THRESH_BINARY)[1] th = cv2.erode(th, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)), iterations = 2) dilated = cv2.dilate(th, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (8, 3)), iterations = 2) image, contours, hier = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) counter = 0 for c in contours: if cv2.contourArea(c) > 500: (x, y, w, h) = cv2.boundingRect(c) cv2.rectangle(frame, (x, y), (x+w, y+h),(0, 255, 0), 1) if firstFrame is True: pedestrains[counter] = Pedestrain(counter, frame, (x, y, w, h)) counter += 1 for i, p in pedestrains.iteritems(): p.update(frame) firstFrame = False frames += 1 cv2.imshow("surveillance", frame) out.write(frame) if cv2.waitKey(110) & 0xff == 27: break out.release() camera.release() if __name__ == '__main__': main()
程序解释
def main():
camera = cv2.VideoCapture(path.join(path.dirname(__file__), "/home/utryjc/Pictures/vtest.avi"))
初始化背景分割器，初始化20针作为影响背景模型的针
history = 20
bs = cv2.createBackgroundSubtractorKNN()
cv2.namedWindow("surveillance")
设置任意字典和firstFrame标志（该标志使得背景分割器能利用这些帧来构建历史） pedestrains = {} firstFrame = True 设立一个帧计数器来更好的识别运动的对象 frames = 0/定义视频编码器 //这里有必要提一下fourcc //FourCC全称Four-Character Codes，代表四字符代码 (four character code), 它是一个32位的标示符，其实就是typedef unsigned int FOURCC;是一种独立标示视频数据流格式的四字符代码。 //因此cv2.VideoWriter_fourcc()函数的作用是输入四个字符代码即可得到对应的视频编码器。 fourcc = cv2.VideoWriter_fourcc(*'XVID')//使用XVID编码器 out = cv2.VideoWriter('output.avi',fourcc, 20.0, (640,480))//出分别是：保存文件名、编码器、帧率、视频宽高fourcc = cv2.VideoWriter_fourcc(*'XVID') out = cv2.VideoWriter('output.avi', fourcc, 20.0, (640, 480)) 一行一行的读取视频帧 while True: print "----------------------FRAME %d------------------" % frames grabbed, frame = camera.read() if (grabbed is False): print "fail to grab frame" break fgmask = bs.apply(frame) if frames < history: frames += 1 continue
cv2.threshold(src, thresh, maxval, type[, dst]) → retval, dst
src：表示的是图片源
thresh：表示的是阈值（起始值）
maxval：表示的是最大值
type：表示的是这里划分的时候使用的是什么类型的算法，常用值为0（cv2.THRESH_BINARY）cv2.threshold(fgmask.copy(), 127, 255, cv2.THRESH_BINARY)[1]#图像阈值处理
th = cv2.erode(th, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)), iterations = 2)#对输入图像用特定结构元素进行腐蚀操作
dilated = cv2.dilate(th, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (8, 3)), iterations = 2) 函数可以对输入图像用特定结构元素进行膨胀操作
image, contours, hier = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
一旦识别到了轮廓，便只是对第一帧中行人的每个轮廓进行实例化，这里对轮廓设置了最小区域，以便能对轮廓进行降噪counter = 0
for c in contours:#每一个行人对象
if cv2.contourArea(c) > 500:
(x, y, w, h) = cv2.boundingRect(c)
cv2.rectangle(frame, (x, y), (x+w, y+h),(0, 255, 0), 1)
if firstFrame is True:
pedestrains[counter] = Pedestrain(counter, frame, (x, y, w, h))
counter += 1
对于每个检测到的行人，都执行update()函数来传递当前帧，这需要在其原来的色彩空间进行，因为行人对象负责绘制它自己的信息
for i, p in pedestrains.iteritems():
这里更新的时候又调用了kalman滤波来预测行人
p.update(frame)
firstFrame设置为False表示不会跟踪更多的行人，只会跟踪现有的行人
firstFrame = False
frames += 1
cv2.imshow("surveillance", frame)
out.write(frame)
if cv2.waitKey(110) & 0xff == 27:
break
out.release()
camera.release()
import cv2 import numpy as np import os.path as path 这个参数主要是在命令行使用 import argparse parser = argparse.ArgumentParser() 假如我们在命令行输入一个文件名加 m或者c,则对应algorithm就为m或者c help用来描述这个选项的作用是选择meanshift还是camshift parser.add_argument("-a", "--algorithm", help = "m (or nothing) for meanShift and c for camshift") 获取属性的值 args = vars(parser.parse_args()) def center(points): x = (points[0][0] + points[1][0] + points[2][0] + points[3][0]) / 4 y = (points[0][1] + points[1][1] + points[2][1] + points[3][1]) / 4 return np.array([np.float32(x), np.float32(y)], np.float32) font = cv2.FONT_HERSHEY_PLAIN创建Pedestrian（行人）类的主要原因是卡尔曼滤波器的性质。卡尔曼滤波器可以通过历史观测来预测对象的位置，然后根据实际数据来矫正预测，但只能对一个对象执行这些操作。因此，每个被跟踪的对象都需要一个卡尔曼滤波器。由此，Pedestrian类会包含卡尔曼滤波器、彩色直方图以及感兴趣区域的信息，这些信息会被CAMShift算法使用。此外，为了获得一些实时信息，需要存储每个行人的ID。 该程序的核心在于背景分割器对象，它能识别感兴趣的区域和与之相对应的运动目标。 当程序启动时，会提取每个区域，实例化Pedestrian类，传递ID，帧以及跟踪窗口的坐标（因此可以提取感兴趣区域（Region of interest，ROI），进而提取ROI的HSV直方图）class Pedestrain(): def __init__(self, id, frame, track_window):#计算给定的ROI直方图，设置kalman滤波器，并将其与对象属性关联 self.id = int(id) x, y, w, h = track_window self.track_window = track_window self.roi = cv2.cvtColor(frame[y:y+h, x:x+w], cv2.COLOR_BGR2HSV) roi_hist = cv2.calcHist([self.roi], [0], None, [16], [0, 180]) self.roi_hist = cv2.normalize(roi_hist, roi_hist, 0, 255, cv2.NORM_MINMAX) self.kalman = cv2.KalmanFilter(4, 2) #cv2.KalmanFilter(4,2)表示转移矩阵维度为4，测量矩阵维度为2self.kalman.measurementMatrix = np.array([[1, 0, 0, 0], [0, 1, 0, 0]], np.float32)#设置测量矩阵 self.kalman.transitionMatrix = np.array([[1, 0, 1, 0], [0, 1, 0, 1], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32)#设置状态矩阵 self.kalman.processNoiseCov = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32) * 0.03#协方差矩阵 self.measurement = np.array((2, 1), np.float32) self.prediction = np.zeros((2, 1), np.float32) self.term_crit = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1) self.center = None self.update(frame) def __del_(self): print "Pedestrain %d destoryed" % self.id def update(self, frame): hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) back_project = cv2.calcBackProject([hsv], [0], self.roi_hist, [0, 180], 1)#计算hsv直方图的反向投影#使用CAMShift来跟踪行人的运动，并根据行人的实际位置校正kalman滤波器 if args.get("algorithm") == "c": ret, self.track_window = cv2.CamShift(back_project, self.track_window, self.term_crit) pts = cv2.boxPoints(ret) pts = np.int0(pts) self.center = center(pts) cv2.polylines(frame, [pts], True, 255, 1)使用均值漂移来跟踪行人的运动，并根据行人的实际位置校正kalman滤波器if not args.get("algorithm") or args.get("algorithm") == "m": ret, self.track_window = cv2.meanShift(back_project, self.track_window, self.term_crit) x, y, w, h = self.track_window self.center = center([[x, y], [x+w, y], [x, y+h], [x+w, y+h]]) cv2.rectangle(frame, (x, y), (x+w, y+h), (255, 255, 0), 2) self.kalman.correct(self.center) prediction = self.kalman.predict() cv2.circle(frame, (int(prediction[0]), int(prediction[1])), 4, (255, 0, 0), -1) 在图的左上角打印行人信息 cv2.putText(frame, "ID:%d -> %s" % (self.id, self.center), (11, (self.id +1)*25+1), font, 0.6, (0, 0, 0), 1, cv2.LINE_AA) cv2.putText(frame, "ID:%d -> %s" % (self.id, self.center), (10, (self.id +1)*25), font, 0.6, (0, 255, 0), 1, cv2.LINE_AA) if __name__ == '__main__': main()
实验效果
参考链接
https://blog.csdn.net/sinat_21258931/article/details/61418681
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
