在Python上使用OpenCV检测和跟踪行人

这是一个跟踪行人的演示程序，由于他们的动作更容易预测，因次行人检测跟踪问题会更容易一些。

该项目使用OpenCV 3.1.0和Python。效果视频网址：https://www.youtube.com/watch?v=ZYgb9e5i_JM

实现Demo如下：

# ====================================================
# ==================== PACKAGES ======================
# ====================================================

## WORKING SPECS:
# OpenCV_3.1.0
# Python_2.7.12

import numpy as np
import cv2
import colorsys
import collections


# ====================================================
# ================== DEFINE CLASS ====================
# ====================================================

class Position(object):
    def __init__(self, _x, _y, _w, _h):
        self.x = _x
        self.y = _y
        self.w = _w
        self.h = _h

    def x(self):
        return self.x

    def y(self):
        return self.y

    def w(self):
        return self.w

    def h(self):
        return self.h

class People(object):
    def __init__(self, _x, _y, _w, _h, _roi, _hue):
        # Position
        self.x = _x
        self.y = _y
        self.w = _w
        self.h = _h
        self.roi = _roi

        # Display of the contour while tracking
        self.hue = _hue
        self.color = hsv2rgb(self.hue%1, 1, 1)

        # Motion Descriptors
        self.center = [_x + _w/2, _y + _h/2]
        self.isIn = checkPosition(boundaryPt1, boundaryPt2, self.center, inCriterion)
        self.isInChangeFrameCount = toleranceCountIOStatus
        self.speed = [0,0]
        self.missingCount = 0

        # ROI - Region of Interest
        self.maxRoi = _roi
        self.roi = _roi

    def x(self):
        return self.x

    def y(self):
        return self.y

    def w(self):
        return self.w

    def h(self):
        return self.h

    def roi(self):
        return self.roi

    def color(self):
        return self.color

    def center(self):
        return self.center

    def maxRoi(self):
        return self.maxRoi

    def isIn(self):
        return self.isIn

    def speed(self):
        return self.speed

    def missingCount(self):
        return self.missingCount

    def isInChangeFrameCount(self):
        return self.isInChangeFrameCount

    def set(self, name, value):
        if name == "x":
            self.x = value
        elif name == "y":
            self.y = value
        elif name == "w":
            self.w = value
        elif name == "h":
            self.h = value
        elif name == "center":
            self.center = value
        elif name == "roi":
            self.roi = value
            # Automatically update maxRoi as roi is updated
            if self.roi.shape[0]*self.roi.shape[1] > self.maxRoi.shape[0]*self.maxRoi.shape[1]:
                self.maxRoi = self.roi
        elif name == "speed":
            self.speed = value
        elif name == "missingCount":
            self.missingCount = value
        elif name == "isIn":
            self.isIn = value
        elif name == "isInChangeFrameCount":
            self.isInChangeFrameCount = value
        else:
            return


# ====================================================
# ===================== FUNCTION =====================
# ====================================================

def averageSize():
    sum = 0
    for i in humanSizeSample:
        sum +=i
    return sum/sampleSize


# Only care about top and bottom
def checkTouchVSide(x, y, w, h, maxW, maxH, tolerance):
    if x <= 0:
        return True
    elif y - tolerance <= 0:
        return True
    elif x + w >= maxW:
        return True
    elif y + h + tolerance >= maxH:
        return True
    else:
        return False


def getExteriorRect(pts):
    xArray = []
    yArray = []
    for pt in pts:
        xArray.append(pt[0])
        yArray.append(pt[1])
    xArray = sorted(xArray)
    yArray = sorted(yArray)
    return (xArray[0], yArray[0], xArray[3] - xArray[0], yArray[3] - yArray[0])


def hsv2rgb(h, s, v):
    return tuple(i * 255 for i in colorsys.hsv_to_rgb(h, s, v))


def checkPosition(boundaryPt1, boundaryPt2, currPos, inCriterion):
    m = (boundaryPt2[1] - boundaryPt1[1])/(boundaryPt2[0] - boundaryPt1[0])
    c = boundaryPt2[1] - m*boundaryPt2[0]
    if inCriterion == "<":
        if currPos[0] * m + c < currPos[1]:
            return True
        else:
            return False
    elif inCriterion == ">":
        if currPos[0] * m + c > currPos[1]:
            return True
        else:
            return False
    else:
        return False

def nothing(x):
    pass

# ====================================================
# ================== VIDEO SOURCE ====================
# ====================================================

srcTest = 'peopleCounter.avi'
srcWebcam = 0
srcMain = '' # live source here
cap = cv2.VideoCapture(srcTest)  # Open video file


# ====================================================
# ================== PRE-CONFIG ======================
# ====================================================

minArea = 500  # default min area to be considered person
maxArea = 4000  # default max area to be considered person
noFrameToCollectSample = 100
toleranceRange = 50  # use for error calculation
toleranceCount = 10  # maximum number of frame an object need to present in order to be accepted
toleranceCountIOStatus = 3  # minimum number of frame between In/Out Status change -> prevent playing with the system
startHue = 0  # In HSV this is RED
hueIncrementValue = 0.1  # increment color every time to differentiate between different people


# ====================================================
# ====================== SETUP =======================
# ====================================================

# fgbg = cv2.createBackgroundSubtractorMOG2(history=500, varThreshold = 16, detectShadows=True)
fgbg = cv2.createBackgroundSubtractorMOG2(detectShadows=True)
sampleSize = 100
humanSizeSample = collections.deque(maxlen=sampleSize)

midHeight = int(cap.get(4) / 2)
maxWidth = cap.get(3)
maxHeight = cap.get(4)

inCriterion = "<"
boundaryPt1 = [0, midHeight-100]
boundaryPt2 = [maxWidth, midHeight]


# ====================================================
# ====================== MAIN ========================
# ====================================================

# Passage Control
allowPassage = True
peopleViolationIn = 0
peopleViolationOut = 0
switch = '0 : PASS \n1 : STOP'

# Controller
cv2.namedWindow('config')
cv2.createTrackbar(switch, 'config', 0, 1, nothing)

# Initializa Other Variable
averageArea = 0.000  # for calculation of min/max size for contour detected
peopleIn = 0  # number of people going up
peopleOut = 0  # number of people going up
frameCounter = 0
maskT = None
passImage = None
detectedPeople = []
detectedContours = []

# take first frame of the video
_ , pFrame = cap.read()

while (cap.isOpened()):

    # Check Passage Status
    status = cv2.getTrackbarPos(switch, 'config')
    if status == 0:
        allowPassage = True
    else:
        allowPassage = False

    # RE-Initialize
    frameInfo = np.zeros((400, 500, 3), np.uint8)
    averageArea = averageSize()
    ret, frame = cap.read()  # read a frame
    frameForView = frame.copy()

    # Clean Frame
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    fgmask = fgbg.apply(gray)
    blur = cv2.medianBlur(fgmask, 5)
    thresh = cv2.threshold(blur, 127, 255, cv2.THRESH_BINARY)[1]  # shadow of MOG@ is grey = 127
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5))
    closing = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel)  # fill any small holes
    opening = cv2.morphologyEx(closing, cv2.MORPH_OPEN, kernel)  # remove noise
    contours = cv2.findContours(opening.copy(), cv2.RETR_EXTERNAL,
                                 cv2.CHAIN_APPROX_SIMPLE)[1]

    mask_opening = cv2.inRange(opening, np.array([0]), np.array([128]))
    noBg = cv2.bitwise_and(frame, frame, mask=mask_opening)

    # Process Contours
    for c in contours:
        # Filter Contour By Size
        if len(humanSizeSample) < 100:
            if cv2.contourArea(c) < minArea or cv2.contourArea(c) > maxArea:
                continue
            else:
                humanSizeSample.append(cv2.contourArea(c))
        else:
            if cv2.contourArea(c) < averageArea/2 or cv2.contourArea(c) > averageArea*3:
                continue
        (x, y, w, h) = cv2.boundingRect(c)
        detectedContours.append(Position(x, y, w, h))

    # Process Detected People
    if len(detectedPeople) != 0:
        for people in detectedPeople:

            # Setup Meanshift/Camshift for Tracking
            track_window = (int(people.x), int(people.y), int(people.w), int(people.h))
            hsv_roi = pOpening[int(people.y):int(people.y) + int(people.h), int(people.x):int(people.x) + int(people.w)]
            mask = cv2.inRange(hsv_roi, np.array(128), np.array(256))
            roi_hist = cv2.calcHist([hsv_roi], [0], mask, [100], [0, 256])
            cv2.normalize(roi_hist, roi_hist, 0, 255, cv2.NORM_MINMAX)
            term_criteria = (cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 1, 1)  # Setup the termination criteria, either 10 iteration or move by atleast 1 pt
            dst = cv2.calcBackProject([opening], [0], roi_hist, [0, 256], 1)
            ret, track_window = cv2.CamShift(dst, track_window, term_criteria)

            # Process POST Tracking
            pts = cv2.boxPoints(ret)
            pts = np.int0(pts)
            img2 = cv2.polylines(frameForView, [pts], True, people.color, 2)
            pos = sum(pts)/len(pts)
            isFound = False
            for dC in detectedContours:
                if dC.x - toleranceRange < pos[0] < dC.x + dC.w + toleranceRange \
                        and dC.y - toleranceRange < pos[1] < dC.y + dC.h + toleranceRange:
                    people.set("x", dC.x)
                    people.set("y", dC.y)
                    people.set("w", dC.w)
                    people.set("h", dC.h)
                    people.set("speed", pos - people.center)
                    people.set("center", pos)
                    people.set("missingCount", 0)
                    detectedContours.remove(dC)
                    isFound = True

                    tR = getExteriorRect(pts)
                    people.set("roi", frame[tR[1]:tR[1]+tR[3], tR[0]:tR[0]+tR[2]])

                    # Process Continuous Tracking
                    prevInStatus = people.isIn
                    currInStatus = checkPosition(boundaryPt1, boundaryPt2, people.center, inCriterion)
                    people.isIn = currInStatus

                    # Check In/Out Status Change
                    if prevInStatus != currInStatus and people.isInChangeFrameCount >= toleranceCountIOStatus:
                        if not allowPassage:
                            passImage = people.roi
                        people.set("isInChangeFrameCount", 0)
                        if currInStatus:
                            peopleIn += 1
                            if not allowPassage:
                                peopleViolationIn += 1
                        else:
                            peopleOut += 1
                            if not allowPassage:
                                peopleViolationOut += 1
                    else:
                        people.set("isInChangeFrameCount", people.isInChangeFrameCount + 1)

            # Process DIS-continuous Tracking
            if not isFound:
                if people.missingCount > toleranceCount:
                    detectedPeople.remove(people)
                else:
                    if checkTouchVSide(people.x + people.speed[0], people.y + people.speed[1], people.w,
                                       people.h, maxWidth, maxHeight, toleranceRange):
                        detectedPeople.remove(people)
                    else:
                        people.set("missingCount", people.missingCount+1)
                        people.set("x", people.x + people.speed[0])
                        people.set("y", people.y + people.speed[1])
                        people.set("center", people.center + people.speed)

    # Check New People
    for dC in detectedContours:
        if checkTouchVSide(dC.x, dC.y, dC.w, dC.h, maxWidth, maxHeight, toleranceRange):
            startHue += hueIncrementValue
            detectedPeople.append(People(dC.x, dC.y, dC.w, dC.h, frame[dC.y:dC.y+dC.h, dC.x:dC.x+dC.w], startHue))

    # RE-set
    detectedContours = []
    pFrame = frame
    pNoBg = noBg
    pOpening = opening
    frameCounter += 1

    # Output
    try:
        # Setup Stats
        textNoOfPeople = "People: " + str(len(detectedPeople))
        textNoIn = "In: " + str(peopleIn)
        textNoOut = "Out: " + str(peopleOut)
        textNoViolationIn = "In: " + str(peopleViolationIn)
        textNoViolationOut = "Out: " + str(peopleViolationOut)

        if allowPassage:
            cv2.line(frameForView, (int(boundaryPt1[0]), int(boundaryPt1[1])),
                     (int(boundaryPt2[0]), int(boundaryPt2[1])), (0, 255, 0), 2)
        else:
            cv2.line(frameForView, (int(boundaryPt1[0]), int(boundaryPt1[1])),
                     (int(boundaryPt2[0]), int(boundaryPt2[1])), (0, 0, 255), 2)

        # Draw Infos
        cv2.putText(frameInfo, textNoOfPeople, (30, 40), cv2.FONT_HERSHEY_SIMPLEX
                    , 1, (255, 255, 255), 1, cv2.LINE_AA)
        cv2.putText(frameInfo, textNoIn, (30, 80), cv2.FONT_HERSHEY_SIMPLEX
                    , 1, (255, 255, 255), 1, cv2.LINE_AA)
        cv2.putText(frameInfo, textNoOut, (30, 120), cv2.FONT_HERSHEY_SIMPLEX
                    , 1, (255, 255, 255), 1, cv2.LINE_AA)
        cv2.line(frameInfo, (0, 160), (640, 160), (255, 255, 255), 1)
        cv2.putText(frameInfo, "VIOLATION", (30, 200), cv2.FONT_HERSHEY_SIMPLEX
                    , 1, (255, 255, 255), 1, cv2.LINE_AA)
        cv2.putText(frameInfo, textNoViolationIn, (30, 240), cv2.FONT_HERSHEY_SIMPLEX
                    , 1, (255, 255, 255), 1, cv2.LINE_AA)
        cv2.putText(frameInfo, textNoViolationOut, (30, 280), cv2.FONT_HERSHEY_SIMPLEX
                    , 1, (255, 255, 255), 1, cv2.LINE_AA)

        # Display
        cv2.imshow('FrameForView', frameForView)
##        cv2.imshow('Frame', frame)
        if passImage != None:
            cv2.imshow('Violators', passImage)
        cv2.imshow('config', frameInfo)

    except Exception as e:
        print(e)
        break

    # Abort and exit with 'Q' or ESC
    k = cv2.waitKey(30) & 0xff
    if k == 27:
        break
    # else:
##    cv2.imwrite(chr(k) + ".jpg", frame)

cap.release()
cv2.destroyAllWindows()