本文为原创文章，转载请注明出处。
本次课题实现目标跟踪一共用到了三个算法，分别是Camshift、Kalman、CSRT，基于Python语言的Tkinter模块实现GUI与接口设计，项目一共包含三个文件：
main.py：
1 # coding:utf-8 2 # 主模块 3 4 5 import Tkinter 6 import tkFileDialog 7 import cv2 8 import time 9 from PIL import ImageTk 10 # 导入自定义模块 11 import track 12 import utils 13 14 15 # 设置窗口800*480 16 root = Tkinter.Tk() 17 root.title("基于视频的实时行人追踪") 18 root.geometry("800x480") 19 20 # 设置背景 21 canvas = Tkinter.Canvas(root, width=800, height=480, highlightthickness=0, borderwidth=0) 22 background_image = ImageTk.PhotoImage(file="background.jpg") # 项目本地路径（背景图片） 23 canvas.create_image(0, 0, anchor="nw", image=background_image) 24 canvas.pack() 25 26 # 显示提示 27 label_a = Tkinter.Label(root, text="基于视频的实时行人追踪", font=("KaiTi", 20), height=2) 28 label_a.pack() 29 canvas.create_window(400, 100, height=25, window=label_a) 30 31 # 显示路径 32 show_path = Tkinter.StringVar() 33 show_path.set("请选择一个文件夹") 34 35 # 显示路径标签 36 label_b = Tkinter.Label(root, textvariable=show_path, font=("Times New Roman", 15), height=2) 37 label_b.pack() 38 canvas.create_window(400, 150, window=label_b) 39 40 # 坐标库 41 ROI = utils.ROI() 42 # 路径库 43 path = utils.Path() 44 45 46 # 选择序列 47 def hit_button_a(): 48 path.init(tkFileDialog.askdirectory(title="Select Folder")) 49 # 显示路径 50 if path.img_path != "": 51 show_path.set("文件路径：" + str(path.img_path)[:-1] + "\n序列总数：" + str(path.sum)) 52 else: 53 show_path.set("路径错误！") 54 55 56 button_a = Tkinter.Button(root, text="选择序列", font=("KaiTi", 15), height=2, command=hit_button_a) 57 button_a.pack() 58 canvas.create_window(400, 200, height=20, window=button_a) 59 60 61 # ROI 62 def hit_button_b(): 63 # 读取首帧图像 64 first_image = cv2.imread(path.pics_list[0]) 65 # ROI 66 ROI.init_window(cv2.selectROI(windowName="ROI", img=first_image, showCrosshair=True, fromCenter=False)) 67 cv2.destroyAllWindows() 68 69 70 button_b = Tkinter.Button(root, text="标记目标", font=("KaiTi", 15), heigh=2, command=hit_button_b) 71 button_b.pack() 72 canvas.create_window(400, 250, height=20, window=button_b) 73 74 75 # 目标追踪 76 77 def hit_button_c(): 78 global camshift, kcf, csrt 79 index = utils.index(path.groundtruth_path) # 读取真值 80 firstframe = True 81 kalman_xy = track.KalmanFilter() 82 kalman_size = track.KalmanFilter() 83 bbox = [0, 0, 0, 0] 84 85 for i in range(0, path.sum): 86 start = time.time() # 开始计时 87 frame = cv2.imread(path.pics_list[i]) # 读取 88 if firstframe: 89 camshift = track.Camshift(frame, ROI.window) 90 kcf = track.KCFtracker(frame, ROI.window) 91 firstframe = False 92 continue 93 # camshift.update(frame) 94 ok = kcf.update(frame) 95 if not ok: 96 mes = (bbox[0], bbox[1], bbox[2], bbox[3]) 97 print mes 98 kcf.tracker.init(frame, mes) 99 ok = kcf.update(frame) 100 end = time.time() # 结束计时 101 seconds = end - start # 处理用时 102 groundtruth = index.groundtruth(i) # 真值 103 window = camshift.window 104 window = kcf.window 105 106 A = window[0] - ROI.window[0] 107 B = window[1] - ROI.window[1] 108 C = window[2] - ROI.window[2] 109 D = window[3] - ROI.window[3] 110 xy = kalman_xy.predict(A, B) 111 size = kalman_size.predict(C, D) # 卡尔曼滤波 112 113 bbox[0] = int(ROI.window[0] + xy[0]) 114 bbox[1] = int(ROI.window[1] + xy[1]) 115 bbox[2] = int(ROI.window[2] + size[0]) 116 bbox[3] = int(ROI.window[3] + size[1]) 117 118 ape = index.APE(bbox, groundtruth) # 像素误差 119 aor = index.AOR(bbox, groundtruth) # 重叠率 120 # 绘制数据曲线 121 # eva.draw(FPS, ape, aor, i) 122 frame = utils.display(seconds, frame, bbox, ape, aor, groundtruth, truth=False) # 跟踪框 123 # 显示 124 cv2.imshow("Track", frame) 125 t = cv2.waitKey(20) & 0xff 126 # 按空格键停止 127 if t == ord(" "): 128 cv2.waitKey(0) 129 # 按ESC键退出 130 if t == 27: 131 cv2.destroyAllWindows() 132 break 133 cv2.destroyAllWindows() 134 print ("跟踪结束!\n") 135 136 137 button_c = Tkinter.Button(root, text="开始追踪", font=("KaiTi", 15), heigh=2, command=hit_button_c) 138 button_c.pack() 139 canvas.create_window(400, 300, height=20, window=button_c) 140 141 142 root.mainloop()
自定义跟踪器模块track.py：
1 # coding:utf-8 2 # 追踪器模块 3 4 5 import cv2 6 import numpy as np 7 8 9 # 得到中心点 10 def center(points): 11 x = (points[0][0] + points[1][0] + points[2][0] + points[3][0]) / 4 12 y = (points[0][1] + points[1][1] + points[2][1] + points[3][1]) / 4 13 return np.array([np.float32(x), np.float32(y)], np.float32) 14 15 16 class Camshift: 17 def __init__(self, frame, ROI): 18 x, y, w, h = ROI 19 self.window = ROI 20 roi = frame[y:y + h, x:x + w] # ROI裁剪 21 hsv_roi = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV) # HSV转换 22 mask = cv2.inRange(hsv_roi, np.array((0., 60., 32.)), np.array((180., 255., 255.))) # 设置阈值 23 self.hist = cv2.calcHist([hsv_roi], [0], mask, [180], [0, 180]) # 直方图 24 cv2.normalize(self.hist, self.hist, 0, 255, cv2.NORM_MINMAX) # 归一化 25 26 def update(self, frame): 27 term_crit = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 1, 10) # 迭代终止标准(最多十次迭代) 28 hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV) # HSV转换 29 dst = cv2.calcBackProject([hsv], [0], self.hist, [0, 180], 1) # 反向投影 30 # cv2.imshow("dst", dst) 31 # cv2.waitKey(10) 32 x, y, w, h = self.window # 跟踪框 33 ret, (x, y, w, h) = cv2.CamShift(dst, (x, y, w, h), term_crit) 34 self.window = (x, y, w, h) 35 36 37 class MILtracker: 38 def __init__(self, frame, ROI): 39 self.window = ROI 40 self.tracker = cv2.TrackerMIL_create() 41 self.tracker.init(frame, self.window) 42 43 def update(self, frame): 44 ok, self.window = self.tracker.update(frame) 45 46 47 class KCFtracker: 48 def __init__(self, frame, ROI): 49 self.window = ROI 50 self.tracker = cv2.TrackerKCF_create() 51 self.tracker = cv2.TrackerCSRT_create() 52 self.tracker.init(frame, self.window) 53 54 def update(self, frame): 55 ok, self.window = self.tracker.update(frame) 56 return ok 57 58 class CSRTtracker: 59 def __init__(self, frame, ROI): 60 self.window = ROI 61 self.tracker = cv2.TrackerCSRT_create() 62 self.tracker.init(frame, self.window) 63 64 def update(self, frame): 65 ok, self.window = self.tracker.update(frame) 66 67 68 class KalmanFilter: 69 def __init__(self): 70 self.kalman = cv2.KalmanFilter(4, 2) 71 self.kalman.measurementMatrix = np.array([[1, 0, 0, 0], [0, 1, 0, 0]], np.float32) 72 self.kalman.transitionMatrix = np.array([[1, 0, 1, 0], [0, 1, 0, 1], [0, 0, 1, 0], [0, 0, 0, 1]], np.float32) 73 self.kalman.processNoiseCov = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]], 74 np.float32) * 0.003 75 self.kalman.measurementNoiseCov = np.array([[1, 0], [0, 1]], np.float32) * 0.01 76 77 def predict(self, x, y): 78 current_mes = np.array([[np.float32(x)], [np.float32(y)]]) 79 self.kalman.correct(current_mes) 80 current_pre = self.kalman.predict() 81 return current_pre
自定义的工具模块utils.py：
1 # coding:utf-8 2 # 工具模块 3 4 5 import cv2 6 import os 7 import re 8 9 10 # 目录存储模块 11 class Path: 12 # 存储文件路径 13 def __init__(self): 14 self.img_path = "" 15 self.groundtruth_path = "" 16 # 目录 17 self.inpics_list = [] 18 # 绝对路径目录 19 self.pics_list = [] 20 self.sum = 0 21 22 # 初始化文件路径 23 def init(self, path): 24 # 请选择包含img和groundtruth的总文件夹 25 if path != '': 26 self.img_path = path + "/img/" 27 self.groundtruth_path = path + "/groundtruth.txt" 28 self.inpics_list = os.listdir(self.img_path) 29 self.inpics_list.sort() 30 # 目录统计 31 self.sum = len(self.inpics_list) 32 # 绝对路径 33 self.pics_list = [self.img_path + x for x in self.inpics_list] 34 35 36 # 坐标储存模块 37 class ROI: 38 # 存储坐标 39 def __init__(self): 40 self.x = 0 41 self.y = 0 42 self.width = 0 43 self.height = 0 44 self.window = [] 45 46 # 单坐标的初始化 47 def init(self, x, y, width, height): 48 self.x = x 49 self.y = y 50 self.width = width 51 self.height = height 52 self.window = (x, y, width, height) 53 54 # 窗口坐标的初始化 55 def init_window(self, window): 56 self.x = window[0] 57 self.y = window[1] 58 self.width = window[2] 59 self.height = window[3] 60 self.window = (window[0], window[1], window[2], window[3]) 61 62 63 # 评价指标模块 64 class index: 65 def __init__(self, path): 66 self.fps = [] 67 self.ape = [] 68 self.aor = [] 69 self.n = [] 70 # 载入真值 71 self.lines = open(path).readlines() 72 73 # 得到真值 74 def groundtruth(self, i): 75 line = [x for x in self.lines] 76 # 切割 77 window = [0, 0, 0, 0] 78 for n in range(0, 4): 79 window[n] = int(re.split("[,\n\t ]", line[i])[n]) 80 return window 81 82 # 像素误差 83 @staticmethod 84 def APE(window, bbox): 85 x1, y1, w1, h1 = window 86 x2, y2, w2, h2 = bbox 87 # 跟踪框中心 88 center = [int(x1 + 1 / 2 * w1), int(y1 + 1 / 2 * h1)] 89 # 真值中心 90 truth_center = [int(x2 + 1 / 2 * w2), int(y2 + 1 / 2 * h2)] 91 # 计算像素误差 92 ape = pow(pow(center[0] - truth_center[0], 2) + pow(center[1] - truth_center[1], 2), .2) 93 ape = round(ape, 2) 94 return ape 95 96 # 重叠率 97 @staticmethod 98 def AOR(window, bbox): 99 x1, y1, w1, h1 = window 100 x2, y2, w2, h2 = bbox 101 col = min(x1 + w1, x2 + w2) - max(x1, x2) 102 row = min(y1 + h1, y2 + h2) - max(y1, y2) 103 intersection = col * row 104 area1 = w1 * h1 105 area2 = w2 * h2 106 coincide = intersection * 1.0 / (area1 + area2 - intersection) * 100 107 aor = round(coincide, 2) 108 return aor 109 110 # 绘制数据曲线 111 def draw(self, fps, ape, aor, number): 112 self.fps.append(fps) 113 self.ape.append(ape) 114 self.aor.append(aor) 115 self.n.append(number) 116 117 118 # 跟踪框显示模块 119 def display(seconds, img, window, ape, aor, groundtruth, truth=False): 120 window = [int(x) for x in window] 121 x, y, w, h = window 122 # 跟踪框 123 img = cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2) 124 if truth: 125 a, b, c, d = groundtruth 126 img = cv2.rectangle(img, (a, b), (a + c, b + d), (0, 255, 0), 2) 127 # 中心点 128 xc = (x + w / 2) 129 yc = (y + h / 2) 130 cv2.circle(img, (xc, yc), 3, (255, 0, 0), -1) 131 # 坐标 132 text = cv2.FONT_HERSHEY_COMPLEX_SMALL 133 size = 1 134 # text = cv2.FONT_ITALIC 135 cv2.putText(img, ('X=' + str(xc)), (10, 20), text, size, (0, 0, 255), 1, cv2.LINE_AA) 136 cv2.putText(img, ('Y=' + str(yc)), (10, 50), text, size, (0, 0, 255), 1, cv2.LINE_AA) 137 # FPS 138 fps = 1 / seconds 139 cv2.putText(img, ('FPS = ' + str(int(fps))), (10, 80), text, size, (0, 255, 0), 1, cv2.LINE_AA) 140 cv2.putText(img, ('APE = ' + str(ape)) + 'pixels', (10, 110), text, size, (0, 255, 255), 1, cv2.LINE_AA) 141 cv2.putText(img, ('AOR = ' + str(aor) + '%'), (10, 140), text, size, (255, 0, 255), 1, cv2.LINE_AA) 142 return img 143 144 def dis(window, img): 145 window = [int(x) for x in window] 146 x, y, w, h = window 147 img = cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2) 148 return img
注：
1.在项目目录下保存一张GUI界面的背景图像background.jpg。
2.在选择样本序列时，格式为：所选定文件夹包含子文件夹img，保存有0001.jpg～...的所有序列，子文件groundtruth.txt真值文件。
3.务必使用低版本（未知原因）的Opencv-contrib，否则不能使用CSRT跟踪器。
TBD.