详解Python人工智能混合高斯模型运动目标检测

本篇内容主要讲解“详解python人工智能混合高斯模型运动目标检测”，感兴趣的朋友不妨来看看。本文介绍的方法操作简单快捷，实用性强。下面就让小编来带大家学习“详解Python人工智能混合高斯模型运动目标检测”吧!

高斯算法提取工作

import cv2import numpy as np# 高斯算法class gaussian:    def __init__(self):        self.mean = np.zeros((1, 3))        self.covariance = 0        self.weight = 0;        self.Next = None        self.Previous = Noneclass node:    def __init__(self):        self.pixel_s = None        self.pixel_r = None        self.no_of_components = 0        self.Next = Noneclass Node1:    def __init__(self):        self.gauss = None        self.no_of_comp = 0        self.Next = Nonecovariance0 = 11.0def Create_gaussian(info1, info2, info3):    ptr = gaussian()    if (ptr is not None):        ptr.mean[1, 1] = info1        ptr.mean[1, 2] = info2        ptr.mean[1, 3] = info3        ptr.covariance = covariance0        ptr.weight = 0.002        ptr.Next = None        ptr.Previous = None    return ptrdef Create_Node(info1, info2, info3):    N_ptr = Node()    if (N_ptr is not None):        N_ptr.Next = None        N_ptr.no_of_components = 1        N_ptr.pixel_s = N_ptr.pixel_r = Create_gaussian(info1, info2, info3)    return N_ptrList_node = []def Insert_End_Node(n):    List_node.append(n)List_gaussian = []def Insert_End_gaussian(n):    List_gaussian.append(n)def Delete_gaussian(n):    List_gaussian.remove(n);class Process:    def __init__(self, alpha, firstFrame):        self.alpha = alpha        self.background = firstFrame    def get_value(self, frame):        self.background = frame * self.alpha + self.background * (1 - self.alpha)        return cv2.absdiff(self.background.astype(np.uint8), frame)def denoise(frame):    frame = cv2.medianBlur(frame, 5)    frame = cv2.GaussianBlur(frame, (5, 5), 0)    return framecapture = cv2.VideoCapture('1.mp4')ret, orig_frame = capture.read( )if ret is True:    value1 = Process(0.1, denoise(orig_frame))    run = Trueelse:    run = Falsewhile (run):    ret, frame = capture.read()    value = False;    if ret is True:        cv2.imshow('input', denoise(frame))        grayscale = value1.get_value(denoise(frame))        ret, mask = cv2.threshold(grayscale, 15, 255, cv2.THRESH_BINARY)        cv2.imshow('mask', mask)        key = cv2.waiTKEy(10) & 0xFF    else:        break    if key == 27:        break    if value == True:        orig_frame = cv2.resize(orig_frame, (340, 260), interpolation=cv2.INTER_CUBIC)        orig_frame = cv2.cvtColor(orig_frame, cv2.COLOR_BGR2GRAY)        orig_image_row = len(orig_frame)        orig_image_col = orig_frame[0]        bin_frame = np.zeros((orig_image_row, orig_image_col))        value = []        for i in range(0, orig_image_row):            for j in range(0, orig_image_col):                N_ptr = Create_Node(orig_frame[i][0], orig_frame[i][1], orig_frame[i][2])                if N_ptr is not None:                    N_ptr.pixel_s.weight = 1.0                    Insert_End_Node(N_ptr)                else:                    print("error")                    exit(0)        nL = orig_image_row        nC = orig_image_col        dell = np.array((1, 3));        mal_dist = 0.0;        temp_cov = 0.0;        alpha = 0.002;        cT = 0.05;        cf = 0.1;        cfbar = 1.0 - cf;        alpha_bar = 1.0 - alpha;        prune = -alpha * cT;        cthr = 0.00001;        var = 0.0        muG = 0.0;        muR = 0.0;        muB = 0.0;        dR = 0.0;        dB = 0.0;        dG = 0.0;        rval = 0.0;        gval = 0.0;        bval = 0.0;        while (1):            duration3 = 0.0;            count = 0;            count1 = 0;            List_node1 = List_node;            counter = 0;            duration = cv2.getTickCount( );            for i in range(0, nL):                r_ptr = orig_frame[i]                b_ptr = bin_frame[i]                for j in range(0, nC):                    sum = 0.0;                    sum1 = 0.0;                    close = False;                    background = 0;                    rval = r_ptr[0][0];                    gval = r_ptr[0][0];                    bval = r_ptr[0][0];                    start = List_node1[counter].pixel_s;                    rear = List_node1[counter].pixel_r;                    ptr = start;                    temp_ptr = None;                    if (List_node1[counter].no_of_component > 4):                        Delete_gaussian(rear);                        List_node1[counter].no_of_component = List_node1[counter].no_of_component - 1;                    for k in range(0, List_node1[counter].no_of_component):                        weight = List_node1[counter].weight;                        mult = alpha / weight;                        weight = weight * alpha_bar + prune;                        if (close == False):                            muR = ptr.mean[0];                            muG = ptr.mean[1];                            muB = ptr.mean[2];                            dR = rval - muR;                            dG = gval - muG;                            dB = bval - muB;                            var = ptr.covariance;                            mal_dist = (dR * dR + dG * dG + dB * dB);                            if ((sum < cfbar) and (mal_dist < 16.0 * var * var)):                                background = 255;                            if (mal_dist < (9.0 * var * var)):                                weight = weight + alpha;                                if mult < 20.0 * alpha:                                    mult = mult;                                else:                                    mult = 20.0 * alpha;                                close = True;                                ptr.mean[0] = muR + mult * dR;                                ptr.mean[1] = muG + mult * dG;                                ptr.mean[2] = muB + mult * dB;                                temp_cov = var + mult * (mal_dist - var);                                if temp_cov < 5.0:                                    ptr.covariance = 5.0                                else:                                    if (temp_cov > 20.0):                                        ptr.covariance = 20.0                                    else:                                        ptr.covariance = temp_cov;                                temp_ptr = ptr;                        if (weight < -prune):                            ptr = Delete_gaussian(ptr);                            weight = 0;                            List_node1[counter].no_of_component = List_node1[counter].no_of_component - 1;                        else:                            sum += weight;                            ptr.weight = weight;                        ptr = ptr.Next;                    if (close == False):                        ptr = gaussian( );                        ptr.weight = alpha;                        ptr.mean[0] = rval;                        ptr.mean[1] = gval;                        ptr.mean[2] = bval;                        ptr.covariance = covariance0;                        ptr.Next = None;                        ptr.Previous = None;                        Insert_End_gaussian(ptr);                        List_gaussian.append(ptr);                        temp_ptr = ptr;                        List_node1[counter].no_of_components = List_node1[counter].no_of_components + 1;                    ptr = start;                    while (ptr != None):                        ptr.weight = ptr.weight / sum;                        ptr = ptr.Next;                    while (temp_ptr != None and temp_ptr.Previous != None):                        if (temp_ptr.weight <= temp_ptr.Previous.weight):                            break;                        else:                            next = temp_ptr.Next;                            previous = temp_ptr.Previous;                            if (start == previous):                                start = temp_ptr;                                previous.Next = next;                                temp_ptr.Previous = previous.Previous;                                temp_ptr.Next = previous;                            if (previous.Previous != None):                                previous.Previous.Next = temp_ptr;                            if (next != None):                                next.Previous = previous;                            else:                                rear = previous;                                previous.Previous = temp_ptr;                        temp_ptr = temp_ptr.Previous;                    List_node1[counter].pixel_s = start;                    List_node1[counter].pixel_r = rear;                    counter = counter + 1;capture.release()cv2.destroyAllwindows()

createBackgroundSubtractORMOG2

• 背景减法 (BS) 是一种常用且广泛使用的技术，用于通过使用静态相机生成前景蒙版（即，包含属于场景中运动物体的像素的二值图像）。

• 顾名思义，BS 计算前景蒙版，在当前帧和背景模型之间执行减法运算，其中包含场景的静态部分，或者更一般地说，根据观察到的场景的特征，可以将所有内容视为背景。
  

背景建模包括两个主要步骤：

• 后台初始化；

• 背景更新。

在第一步中，计算背景的初始模型，而在第二步中，更新该模型以适应场景中可能的变化。

import cv2#构造VideoCapture对象cap = cv2.VideoCapture('1.mp4')# 创建一个背景分割器# createBackgroundSubtractorMOG2()函数里，可以指定detectShadows的值# detectShadows=True，表示检测阴影，反之不检测阴影。默认是truefgbg  = cv2.createBackgroundSubtractorMOG2()while True :    ret, frame = cap.read() # 读取视频    fgmask = fgbg.apply(frame) # 背景分割    cv2.imshow('frame', fgmask) # 显示分割结果    if cv2.waitKey(100) & 0xff == ord('q'):        breakcap.release()cv2.destroyAllWindows()

到此，相信大家对“详解Python人工智能混合高斯模型运动目标检测”有了更深的了解，不妨来实际操作一番吧！这里是编程网网站，更多相关内容可以进入相关频道进行查询，关注我们，继续学习！