第八章 目标跟踪
1检测目标的移动
基本的运动检测,示例代码如下:
import cv2 import numpy as np # 捕获摄像头图像 camera = cv2.VideoCapture(0) # es = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (10, 10)) kernel = np.ones((5, 5), np.uint8) background = None while (True): ret, frame = camera.read() # 将第一帧设为图像的背景 if background is None: # 转换颜色空间 background = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) # 高斯模糊 background = cv2.GaussianBlur(background, (21, 21), 0) continue # 转换颜色空间并作模糊处理 gray_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) gray_frame = cv2.GaussianBlur(gray_frame, (21, 21), 0) # 取得差分图 diff = cv2.absdiff(background, gray_frame) diff = cv2.threshold(diff, 25, 255, cv2.THRESH_BINARY)[1] # 膨胀 diff = cv2.dilate(diff, es, iterations=2) # 得到图像中目标的轮廓 image, cnts, hierarchy = cv2.findContours(diff.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) for c in cnts: if cv2.contourArea(c) < 1500: continue # 计算矩形边框 (x, y, w, h) = cv2.boundingRect(c) # 绘制矩形 cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 2) # 显示图像 cv2.imshow('contours', frame) cv2.imshow('dif', diff) if cv2.waitKey(int(1000 / 12)) & 0xFF == ord('q'): break cv2.destroyAllWindows() camera.release()
运行结果如下:
2背景分割器 knn mog2和GMG
Opencv3有三种背景分割器
K-nearest(knn)
Mixture of Gaussians(MOG2)
Geometric multigid(GMC)
backgroundSubtractor用于分割前景和背景
示例代码如下:
import cv2 import numpy as np cv2.ocl.setUseOpenCL(False) cap = cv2.VideoCapture(0) mog = cv2.createBackgroundSubtractorMOG2() while (True): ret, frame = cap.read() fgmask = mog.apply(frame) cv2.imshow('frame', fgmask) if cv2.waitKey(30) & 0xFF == ord('q'): break cap.release() cv2.destroyAllWindows()
运行结果如下:
使用backgroundSubtractorKNN来实现运动检测
示例代码如下:
import cv2 cv2.ocl.setUseOpenCL(False) bs = cv2.createBackgroundSubtractorKNN(detectShadows=True) # 读取本地视频 camera = cv2.VideoCapture('../traffic.flv') while (True): ret, frame = camera.read() fgmask = bs.apply(frame.copy()) # 设置阈值 th = cv2.threshold(fgmask, # 源图像 244, # 阈值 255, # 最大值 cv2.THRESH_BINARY)[1] # 阈值类型 # 膨胀 dilated = cv2.dilate(th, # 源图像 cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)), # 内核 iterations=2) # 腐蚀次数 # 查找图像中的目标轮廓 image, contours, hier = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) for c in contours: if cv2.contourArea(c) > 1600: (x, y, w, h) = cv2.boundingRect(c) cv2.rectangle(frame, (x, y), (x + w, y + h), (255, 255, 0), 2) cv2.imshow('mog', fgmask) # 分割前景与背景 cv2.imshow('thresh', th) # cv2.imshow('detection', frame) # 运动检测结果 if cv2.waitKey(30) & 0xFF == 27: break camera.release() cv2.destroyAllWindows()
运行结果如下:
均值漂移meanShift
示例代码如下:
import cv2 import numpy as np # 取得摄像头图像 cap = cv2.VideoCapture(0) ret, frame = cap.read() # 设置跟踪窗体大小 r, h, c, w = 10, 200, 10, 200 track_window = (c, r, w, h) # 提取roi roi = frame[r:r + h, c:c + w] # 转换颜色空间 hsv_roi = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # 根据阈值构建掩码 mask = cv2.inRange(hsv_roi, np.array((100., 30., 32.)), np.array((180., 120., 255.))) # 计算roi图形的彩色直方图 roi_hist = cv2.calcHist([hsv_roi], [0], mask, [180], [0, 180]) cv2.normalize(roi_hist, roi_hist, 0, 255, cv2.NORM_MINMAX) # 指定停止条件 term_crit = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1) while (True): ret, frame = cap.read() if ret == True: # 更换颜色空间 hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # histogram back projection calculation 直方图反向投影 dst = cv2.calcBackProject([hsv], [0], roi_hist, [0, 180], 1) # 均值漂移 ret, track_window = cv2.meanShift(dst, track_window, term_crit) # 绘制矩形显示图像 x, y, w, h = track_window img2 = cv2.rectangle(frame, (x, y), (x + w, y + h), 255, 2) cv2.imshow('img2', img2) # esc退出 if cv2.waitKey(60) & 0xFF == 27: break else: break cv2.destroyAllWindows() cap.release()
运行结果如下:
彩色直方图
calHist函数
函数原型:
def calcHist(images, #源图像
channels, #通道列表
mask,#可选的掩码
histSize, #每个维度下直方图数组的大小
ranges,#每一个维度下直方图bin的上下界的数组
hist=None,#输出直方图是一个[]维稠密度的数组
accumulate=None)#累计标志
Camshift
示例代码如下:
#!/usr/bin/env python # -*- coding: utf-8 -*- # @Time : 2016/12/15 16:48 # @Author : Retacn # @Site : camshift实现物体跟踪 # @File : camshift.py # @Software: PyCharm __author__ = "retacn" __copyright__ = "property of mankind." __license__ = "CN" __version__ = "0.0.1" __maintainer__ = "retacn" __email__ = "zhenhuayue@sina.com" __status__ = "Development" import cv2 import numpy as np # 取得摄像头图像 cap = cv2.VideoCapture(0) ret, frame = cap.read() # 设置跟踪窗体大小 r, h, c, w = 300, 200, 400, 300 track_window = (c, r, w, h) # 提取roi roi = frame[r:r + h, c:c + w] # 转换颜色空间 hsv_roi = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # 根据阈值构建掩码 mask = cv2.inRange(hsv_roi, np.array((100., 30., 32.)), np.array((180., 120., 255.))) # 计算roi图形的彩色直方图 roi_hist = cv2.calcHist([hsv_roi], [0], mask, [180], [0, 180]) cv2.normalize(roi_hist, roi_hist, 0, 255, cv2.NORM_MINMAX) # 指定停止条件 term_crit = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1) while (True): ret, frame = cap.read() if ret == True: # 更换颜色空间 hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # histogram back projection calculation 直方图反向投影 dst = cv2.calcBackProject([hsv], [0], roi_hist, [0, 180], 1) # 均值漂移 ret, track_window = cv2.CamShift(dst, track_window, term_crit) # 绘制矩形显示图像 pts = cv2.boxPoints(ret) pts = np.int0(pts) img2 = cv2.polylines(frame, [pts], True, 255, 2) cv2.imshow('img2', img2) # esc退出 if cv2.waitKey(60) & 0xFF == 27: break else: break cv2.destroyAllWindows() cap.release()
运行结果如下:
4 卡尔曼滤波器
函数原型为：
def KalmanFilter(dynamParams=None,#状态的维度
measureParams=None, #测量的维度
controlParams=None,#控制的维度
type=None)#矩阵的类型
示例代码如下：
import cv2 import numpy as np # 创建空帧 frame = np.zeros((800, 800, 3), np.uint8) # 测量坐标 last_measurement = current_measurement = np.array((2, 1), np.float32) # 鼠标运动预测 last_prediction = current_predication = np.zeros((2, 1), np.float32) def mousemove(event, x, y, s, p): # 设置全局变量 global frame, measurements, current_measurement, last_measurement, current_predication, last_prediction last_prediction = current_predication last_measurement = current_measurement current_measurement = np.array([[np.float32(x)], [np.float32(y)]]) kalman.correct(current_measurement) current_predication = kalman.predict() # 实际移动起始点 lmx, lmy = last_measurement[0], last_measurement[1] cmx, cmy = current_measurement[0], current_measurement[1] # 预测线起止点 lpx, lpy = last_prediction[0], last_prediction[1] cpx, cpy = current_predication[0], current_predication[1] # 绘制连线 cv2.line(frame, (lmx, lmy), (cmx, cmy), (0, 100, 0)) # 绿色 cv2.line(frame, (lpx, lpy), (cpx, cpy), (0, 0, 200)) # 红色 # 创建窗体 cv2.namedWindow('mouse_detection') # 注册鼠标事件的回调函数 cv2.setMouseCallback('mouse_detection', mousemove) # 卡尔曼滤波器 kalman = cv2.KalmanFilter(4, 2) kalman.measurementMatrix = np.array([[1, 0, 0, 0], [0, 1, 0, 0]], np.float32) kalman.transitionMatrix = np.array([[1, 0, 1, 0], [0, 1, 0, 1], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32) kalman.processNoiseCov = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32) * 0.03 while (True): cv2.imshow('mouse_detection', frame) if cv2.waitKey(30) & 0xFF == 27: break cv2.destroyAllWindows()
运行结果如下：
一个基于行人跟踪的例子
示例代码如下:
import cv2 import numpy as np import os.path as path import argparse font = cv2.FONT_HERSHEY_SIMPLEX parser = argparse.ArgumentParser() parser.add_argument("-a", "--algorithm", help="m (or nothing) for meanShift and c for camshift") args = vars(parser.parse_args()) # 计算矩阵中心(行人位置) 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 # print(np.array([np.float32(x), np.float32(y)], np.float32)) # [ 588. 257.5] return np.array([np.float32(x), np.float32(y)], np.float32) # 行人 class Pedestrian(): def __init__(self, id, frame, track_window): self.id = int(id) # 行人id x, y, w, h = track_window # 跟踪窗体 self.track_window = track_window # 更换颜色空间 self.roi = cv2.cvtColor(frame[y:y + h, x:x + w], cv2.COLOR_BGR2HSV) # 计算roi图形的彩色直方图 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) # 设置卡尔曼滤波器 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 # 测量坐标 self.measurement = np.array((2, 1), np.float32) # 鼠标运动预测 self.predication = np.zeros((2, 1), np.float32) # 指定停止条件 self.term_crit = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 10, 1) self.center = None self.update(frame) def __del__(self): print('Pedestrian %d destroyed' % self.id) # 更新图像帧 def update(self, frame): # 更换颜色空间 hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # histogram back projection calculation 直方图反向投影 back_project = cv2.calcBackProject([hsv], [0], self.roi_hist, [0, 180], 1) # camshift if args.get('algorithm') == 'c': ret, self.track_window = cv2.CamShift(back_project, self.track_window, self.term_crit) # 绘制跟踪框 pts = cv2.boxPoints(ret) pts = np.int0(pts) self.center = center(pts) cv2.polylines(frame, [pts], True, 255, 1) # 均值漂移 if not args.get('algorithm') or args.get('algorithm') == 'm': ret, self.track_window = cv2.meanShift(back_project, self.track_window, self.term_crit) # 绘制跟踪框 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, (0, 255, 0), -1) # 计数器 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) # 跟踪窗口坐标 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(): # 加载视频 # camera = cv2.VideoCapture('../movie.mpg') # camera = cv2.VideoCapture('../traffic.flv') camera = cv2.VideoCapture('../768x576.avi') # 初始化背景分割器 history = 20 bs = cv2.createBackgroundSubtractorKNN(detectShadows=True) # 创建显示主窗口 cv2.namedWindow('surveillance') pedestrians = {} # 行人字典 firstFrame = True frames = 0 fourcc = cv2.VideoWriter_fourcc(*'XVID') out = cv2.VideoWriter('../output.avi', fourcc, 20.0, (640, 480)) while (True): print('----------------------frmae %d----------------' % frames) grabbed, frane = camera.read() if (grabbed is False): print("failed to grab frame") break ret, frame = camera.read() fgmask = bs.apply(frame) if frames < history: frames += 1 continue # 设置阈值 th = cv2.threshold(fgmask.copy(), 127, 255, cv2.THRESH_BINARY)[1] # 腐蚀 th = cv2.erode(th, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3)), iterations=2) # 膨胀 dilated = cv2.dilate(th, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (8, 3)), iterations=2) # 查找轮廓 image, contours, hier = cv2.findContours(dilated, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) counter = 0 for c in contours: if cv2.contourArea(c) > 500: # 边界数组 (x, y, w, h) = cv2.boundingRect(c) # 绘制矩形 cv2.rectangle(frame, (x, y), (x + w, y + h), (0, 255, 0), 1) if firstFrame is True: pedestrians[counter] = Pedestrian(counter, frame, (x, y, w, h)) counter += 1 # 更新帧内容 for i, p in pedestrians.items(): p.update(frame) # false 只跟踪已有的行人 # firstFrame = True firstFrame = False frames += 1 # 显示 cv2.imshow('surveillance', frame) out.write(frame) if cv2.waitKey(120) & 0xFF == 27: # esc退出 break out.release() camera.release() if __name__ == "__main__": main()
运行结果如下: