python如何实现单目标、多目标、多尺度、自定义特征的KCF跟踪算法-创新互联
小编给大家分享一下python如何实现单目标、多目标、多尺度、自定义特征的KCF跟踪算法,相信大部分人都还不怎么了解,因此分享这篇文章给大家参考一下,希望大家阅读完这篇文章后大有收获,下面让我们一起去了解一下吧!
创新互联专注于成都网站制作、网站建设、外贸网站建设、网页设计、网站制作、网站开发。公司秉持“客户至上,用心服务”的宗旨,从客户的利益和观点出发,让客户在网络营销中找到自己的驻足之地。尊重和关怀每一位客户,用严谨的态度对待客户,用专业的服务创造价值,成为客户值得信赖的朋友,为客户解除后顾之忧。单目标跟踪:
直接调用opencv中封装的tracker即可。
#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ Created on Sun Jan 5 17:50:47 2020 第四章 kcf跟踪 @author: youxinlin """ import cv2 from items import MessageItem import time import numpy as np ''' 监视者模块,负责入侵检测,目标跟踪 ''' class WatchDog(object): #入侵检测者模块,用于入侵检测 def __init__(self,frame=None): #运动检测器构造函数 self._background = None if frame is not None: self._background = cv2.GaussianBlur(cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY),(21,21),0) self.es = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (10, 10)) def isWorking(self): #运动检测器是否工作 return self._background is not None def startWorking(self,frame): #运动检测器开始工作 if frame is not None: self._background = cv2.GaussianBlur(cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY), (21, 21), 0) def stopWorking(self): #运动检测器结束工作 self._background = None def analyze(self,frame): #运动检测 if frame is None or self._background is None: return sample_frame = cv2.GaussianBlur(cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY),(21,21),0) diff = cv2.absdiff(self._background,sample_frame) diff = cv2.threshold(diff, 25, 255, cv2.THRESH_BINARY)[1] diff = cv2.dilate(diff, self.es, iterations=2) image, cnts, hierarchy = cv2.findContours(diff.copy(),cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) coordinate = [] bigC = None bigMulti = 0 for c in cnts: if cv2.contourArea(c) < 1500: continue (x,y,w,h) = cv2.boundingRect(c) if w * h > bigMulti: bigMulti = w * h bigC = ((x,y),(x+w,y+h)) if bigC: cv2.rectangle(frame, bigC[0],bigC[1], (255,0,0), 2, 1) coordinate.append(bigC) message = {"coord":coordinate} message['msg'] = None return MessageItem(frame,message) class Tracker(object): ''' 追踪者模块,用于追踪指定目标 ''' def __init__(self,tracker_type = "BOOSTING",draw_coord = True): ''' 初始化追踪器种类 ''' #获得opencv版本 (major_ver, minor_ver, subminor_ver) = (cv2.__version__).split('.') self.tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'GOTURN'] self.tracker_type = tracker_type self.isWorking = False self.draw_coord = draw_coord #构造追踪器 if int(minor_ver) < 3: self.tracker = cv2.Tracker_create(tracker_type) else: if tracker_type == 'BOOSTING': self.tracker = cv2.TrackerBoosting_create() if tracker_type == 'MIL': self.tracker = cv2.TrackerMIL_create() if tracker_type == 'KCF': self.tracker = cv2.TrackerKCF_create() if tracker_type == 'TLD': self.tracker = cv2.TrackerTLD_create() if tracker_type == 'MEDIANFLOW': self.tracker = cv2.TrackerMedianFlow_create() if tracker_type == 'GOTURN': self.tracker = cv2.TrackerGOTURN_create() def initWorking(self,frame,box): ''' 追踪器工作初始化 frame:初始化追踪画面 box:追踪的区域 ''' if not self.tracker: raise Exception("追踪器未初始化") status = self.tracker.init(frame,box) if not status: raise Exception("追踪器工作初始化失败") self.coord = box self.isWorking = True def track(self,frame): ''' 开启追踪 ''' message = None if self.isWorking: status,self.coord = self.tracker.update(frame) if status: message = {"coord":[((int(self.coord[0]), int(self.coord[1])),(int(self.coord[0] + self.coord[2]), int(self.coord[1] + self.coord[3])))]} if self.draw_coord: p1 = (int(self.coord[0]), int(self.coord[1])) p2 = (int(self.coord[0] + self.coord[2]), int(self.coord[1] + self.coord[3])) cv2.rectangle(frame, p1, p2, (255,0,0), 2, 1) message['msg'] = "is tracking" return MessageItem(frame,message) class ObjectTracker(object): def __init__(self,dataSet): self.cascade = cv2.CascadeClassifier(dataSet) def track(self,frame): gray = cv2.cvtColor(frame,cv2.COLOR_BGR2GRAY) faces = self.cascade.detectMultiScale(gray,1.03,5) for (x,y,w,h) in faces: cv2.rectangle(frame,(x,y),(x+w,y+h),(0,0,255),2) return frame if __name__ == '__main__' : # tracker_types = ['BOOSTING', 'MIL','KCF', 'TLD', 'MEDIANFLOW', 'GOTURN'] tracker = Tracker(tracker_type="KCF") # video = cv2.VideoCapture(0) # video = cv2.VideoCapture("complex1.mov") video = cv2.VideoCapture(r"/Users/youxinlin/Desktop/video_data/complex1.MOV") ok, frame = video.read() bbox = cv2.selectROI(frame, False) tracker.initWorking(frame,bbox) while True: _,frame = video.read(); if(_): item = tracker.track(frame); cv2.imshow("track",item.getFrame()) k = cv2.waitKey(1) & 0xff if k == 27: break
附带items.py,放在同个文件夹下:
#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ Created on Sun Jan 5 17:51:04 2020 @author: youxinlin """ import json from utils import IOUtil ''' 信息封装类 ''' class MessageItem(object): #用于封装信息的类,包含图片和其他信息 def __init__(self,frame,message): self._frame = frame self._message = message def getFrame(self): #图片信息 return self._frame def getMessage(self): #文字信息,json格式 return self._message def getBase64Frame(self): #返回base64格式的图片,将BGR图像转化为RGB图像 jepg = IOUtil.array_to_bytes(self._frame[...,::-1]) return IOUtil.bytes_to_base64(jepg) def getBase64FrameByte(self): #返回base64格式图片的bytes return bytes(self.getBase64Frame()) def getJson(self): #获得json数据格式 dicdata = {"frame":self.getBase64Frame().decode(),"message":self.getMessage()} return json.dumps(dicdata) def getBinaryFrame(self): return IOUtil.array_to_bytes(self._frame[...,::-1])
utils.py:也放在同一个文件夹下。
#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ Created on Sun Jan 5 17:51:40 2020 @author: youxinlin """ import time import numpy import base64 import os import logging import sys from PIL import Image from io import BytesIO #工具类 class IOUtil(object): #流操作工具类 @staticmethod def array_to_bytes(pic,formatter="jpeg",quality=70): ''' 静态方法,将numpy数组转化二进制流 :param pic: numpy数组 :param format: 图片格式 :param quality:压缩比,压缩比越高,产生的二进制数据越短 :return: ''' stream = BytesIO() picture = Image.fromarray(pic) picture.save(stream,format=formatter,quality=quality) jepg = stream.getvalue() stream.close() return jepg @staticmethod def bytes_to_base64(byte): ''' 静态方法,bytes转base64编码 :param byte: :return: ''' return base64.b64encode(byte) @staticmethod def transport_rgb(frame): ''' 将bgr图像转化为rgb图像,或者将rgb图像转化为bgr图像 ''' return frame[...,::-1] @staticmethod def byte_to_package(bytes,cmd,var=1): ''' 将每一帧的图片流的二进制数据进行分包 :param byte: 二进制文件 :param cmd:命令 :return: ''' head = [ver,len(byte),cmd] headPack = struct.pack("!3I", *head) senddata = headPack+byte return senddata @staticmethod def mkdir(filePath): ''' 创建文件夹 ''' if not os.path.exists(filePath): os.mkdir(filePath) @staticmethod def countCenter(box): ''' 计算一个矩形的中心 ''' return (int(abs(box[0][0] - box[1][0])*0.5) + box[0][0],int(abs(box[0][1] - box[1][1])*0.5) +box[0][1]) @staticmethod def countBox(center): ''' 根据两个点计算出,x,y,c,r ''' return (center[0][0],center[0][1],center[1][0]-center[0][0],center[1][1]-center[0][1]) @staticmethod def getImageFileName(): return time.strftime("%Y_%m_%d_%H_%M_%S", time.localtime())+'.png'
多目标跟踪:
和单目标差不多,改用MultiTracker_create()
#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ Created on Sun Jan 5 18:02:33 2020
多目标跟踪
@author: youxinlin """import numpy as np import cv2 import sys ''' if len(sys.argv) != 2: print('Input video name is missing') exit() ''' print('Select multiple tracking targets') cv2.namedWindow("tracking") camera = cv2.VideoCapture(r"/Users/youxinlin/Desktop/video_data/complex6.MOV") #camera = cv2.VideoCapture(0) tracker = cv2.MultiTracker_create() #多目标跟踪 a= cv2.Tracker_c init_once = False ok, image=camera.read() if not ok: print('Failed to read video') exit() bbox1 = cv2.selectROI('tracking', image) bbox2 = cv2.selectROI('tracking', image) bbox3 = cv2.selectROI('tracking', image) while camera.isOpened(): ok, image=camera.read() if not ok: print ('no image to read') break if not init_once: ok = tracker.add(cv2.TrackerKCF_create(),image,bbox1) ok = tracker.add(cv2.TrackerKCF_create( ),image, bbox2) ok = tracker.add(cv2.TrackerKCF_create(),image, bbox3) init_once = True ok, boxes = tracker.update(image) for newbox in boxes: p1 = (int(newbox[0]), int(newbox[1])) p2 = (int(newbox[0] + newbox[2]), int(newbox[1] + newbox[3])) cv2.rectangle(image, p1, p2, (0,0,255)) cv2.imshow('tracking', image) k = cv2.waitKey(1) if k == 27 : break # esc pressed
多尺度检测的KCF、自定义所用特征的KCF
在一些场景下,不想使用默认的hog特征跟踪,或需要对比不同特征的跟踪效果,那么封装好的方法似乎不可用,需要可以自己撸一波kcf的代码,从而使用自己设定的特征。
以上是“python如何实现单目标、多目标、多尺度、自定义特征的KCF跟踪算法”这篇文章的所有内容,感谢各位的阅读!相信大家都有了一定的了解,希望分享的内容对大家有所帮助,如果还想学习更多知识,欢迎关注创新互联行业资讯频道!
网页名称:python如何实现单目标、多目标、多尺度、自定义特征的KCF跟踪算法-创新互联
文章出自:http://azwzsj.com/article/ihsci.html