社交距离检测——目标检测+距离检测

参考

window10下打开摄像头实现Pytorch-YOLOv3的实时监测
python+OpenCV+YOLOv3打开笔记本摄像头模型检测
基于yoloV3的目标检测
社交距离检测器——Tensorflow检测模型设计

1.社交距离的目标检测，只需要检测人即可

import numpy as np
import cv2


def video_demo():
    # 加载已经训练好的模型路径，可以是绝对路径或者相对路径
    weightsPath = r'D:\YOLO\darknet-master\build\darknet\x64\yolov3.weights'
    configPath = r"D:\YOLO\darknet-master\cfg\yolov3.cfg"
    labelsPath = r"D:\YOLO\darknet-master\data\coco.names"
    # 初始化一些参数
    LABELS = open(labelsPath).read().strip().split("\n")  # 物体类别
    COLORS = np.random.randint(0, 255, size=(len(LABELS), 3), dtype="uint8")  # 颜色
    net = cv2.dnn.readNetFromDarknet(configPath, weightsPath)
    # net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
    # net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)

    # 读入待检测的图像
    # 0是代表摄像头编号，只有一个的话默认为0
    # capture = cv2.VideoCapture(0)
    capture = cv2.VideoCapture(r'C:\Users\lenovo\Desktop\shipin\01.mp4')
    # 读入待检测的图像
    # 0是代表摄像头编号，只有一个的话默认为0
    yolo_num = 1
    while (True):
        boxes = []
        confidences = []
        classIDs = []
        print("yolo%d" % yolo_num)
        yolo_num = yolo_num + 1
        ref, image = capture.read()
        #image = cv2.resize(image, (300, 300), fx=0.25, fy=0.25)
        (H, W) = image.shape[:2]

        # 得到 YOLO需要的输出层
        ln = net.getLayerNames()
        ln = [ln[i[0] - 1] for i in net.getUnconnectedOutLayers()]

        # 从输入图像构造一个blob，然后通过加载的模型，给我们提供边界框和相关概率
        blob = cv2.dnn.blobFromImage(image, 1 / 255.0, (416, 416), swapRB=True, crop=False)
        net.setInput(blob)
        layerOutputs = net.forward(ln)

        # 在每层输出上循环
        for output in layerOutputs:
            # 对每个检测进行循环
            for detection in output:
                scores = detection[5:]
                classID = np.argmax(scores)
                confidence = scores[classID]
                # 过滤掉那些置信度较小的检测结果
                if confidence > 0.5:
                    # 框后接框的宽度和高度
                    box = detection[0:4] * np.array([W, H, W, H])
                    (centerX, centerY, width, height) = box.astype("int")
                    # 边框的左上角
                    x = int(centerX - (width / 2))
                    y = int(centerY - (height / 2))
                    # 更新检测出来的框
                    boxes.append([x, y, int(width), int(height)])
                    confidences.append(float(confidence))
                    classIDs.append(classID)
        # 极大值抑制
        idxs = cv2.dnn.NMSBoxes(boxes, confidences, 0.2, 0.3)
        if len(idxs) > 0:
            for i in idxs.flatten():
                (x, y) = (boxes[i][0], boxes[i][1])
                (w, h) = (boxes[i][2], boxes[i][3])
                # 在原图上绘制边框和类别
                color = [int(c) for c in COLORS[classIDs[i]]]
                if LABELS[classIDs[i]] == "person":  ###只检测人
                    print(" Label: %s:(x,y,w,h)=(%d,%d,%d,%d)" % (LABELS[classIDs[i]], x, y, w, h))
                    cv2.rectangle(image, (x, y), (x + w, y + h), color, 2)
                    text = '{}: {:.3f}'.format(LABELS[classIDs[i]], confidences[i])
                    (text_w, text_h), baseline = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 2)
                    cv2.rectangle(image, (x, y - text_h - baseline), (x + text_w, y), color, -1)
                    cv2.putText(image, text, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 2)
                    #cv2.rectangle(image, (x, y), (x + w, y + h), color, 2)  ##左上，右下
        cv2.imshow("Image", image)
        c = cv2.waitKey(1) & 0xff
        if c == 27:
            capture.release()
            break
    cv2.destroyAllWindows()

video_demo()

2.距离检测

根据框的质心或者框的底部中心点，检测两点之间的距离。

                if math.sqrt(math.pow((pair[0][0] - pair[1][0]), 2) + math.pow((pair[0][1] - pair[1][1]), 2)) < int(distance_minimum) :
                    index_pt1 = list_indexes[i][0]
                    index_pt2 = list_indexes[i][1]
                    change_color_originalframe(index_pt1,index_pt2)

3.完整代码

import numpy as np
import cv2
import itertools
import math

def video_demo():
    # 加载已经训练好的模型路径，可以是绝对路径或者相对路径
    weightsPath = r'D:\YOLO\darknet-master\build\darknet\x64\yolov3.weights'
    configPath = r"D:\YOLO\darknet-master\cfg\yolov3.cfg"
    labelsPath = r"D:\YOLO\darknet-master\data\coco.names"
    # 初始化一些参数
    LABELS = open(labelsPath).read().strip().split("\n")  # 物体类别
    #COLORS = np.random.randint(0, 255, size=(len(LABELS), 3), dtype="uint8")  # 颜色
    net = cv2.dnn.readNetFromDarknet(configPath, weightsPath)
    # net.setPreferableBackend(cv2.dnn.DNN_BACKEND_CUDA)
    # net.setPreferableTarget(cv2.dnn.DNN_TARGET_CUDA)

    capture = cv2.VideoCapture(r'C:/Users/lenovo/Desktop/shipin/test.mp4')
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    out = cv2.VideoWriter(r'C:\Users\lenovo\Desktop\shipin\camera_test2.mp4', fourcc, 24.0, (768, 576))
    yolo_num = 1
    while (True):
        boxes = []
        confidences = []
        classIDs = []
        print("yolo%d" % yolo_num)
        yolo_num = yolo_num + 1
        a = 1
        ret, image = capture.read()
        if ret is False:
            break
        (H, W) = image.shape[:2]  #图片的大小

        # 得到 YOLO需要的输出层
        ln = net.getLayerNames()
        ln = [ln[i[0] - 1] for i in net.getUnconnectedOutLayers()]

        # 从输入图像构造一个blob，然后通过加载的模型，给我们提供边界框和相关概率
        blob = cv2.dnn.blobFromImage(image, 1 / 255.0, (416, 416), swapRB=True, crop=False)
        net.setInput(blob)
        layerOutputs = net.forward(ln)

        # 在每层输出上循环
        for output in layerOutputs:
            # 对每个检测进行循环
            for detection in output:
                scores = detection[5:]
                classID = np.argmax(scores)
                confidence = scores[classID]
                # 过滤掉那些置信度较小的检测结果
                if confidence > 0.5:
                    # 框后接框的宽度和高度
                    box = detection[0:4] * np.array([W, H, W, H])
                    (centerX, centerY, width, height) = box.astype("int")
                    # 边框的左上角
                    x = int(centerX - (width / 2))
                    y = int(centerY - (height / 2))
                    # 更新检测出来的框
                    boxes.append([x, y, int(width), int(height)])
                    confidences.append(float(confidence))
                    classIDs.append(classID)

        # 极大值抑制
        idxs = cv2.dnn.NMSBoxes(boxes, confidences, 0.2, 0.3)
        if len(idxs) > 0:
            transformed_downoids = []
            transformed_g_list = []
            for i in idxs.flatten():
                (x, y) = (boxes[i][0], boxes[i][1])
                (w, h) = (boxes[i][2], boxes[i][3])
                if LABELS[classIDs[i]] == "person":  ###只检测人
                    print(" Label: %s:(x,y,w,h,a)=(%d,%d,%d,%d,%d)" % (LABELS[classIDs[i]], x, y, w, h, a))
                    #print(" Label: %s:(x,y,w,h)=(%d,%d,%d,%d)" % (LABELS[classIDs[i]], x, y, w, h))
                    a = a + 1
                    cv2.rectangle(image, (x, y), (x + w, y + h), (0,255,0), 2)
                    text = '{}: {:.3f}'.format(LABELS[classIDs[i]], confidences[i])
                    (text_w, text_h), baseline = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 2)
                    cv2.rectangle(image, (x, y - text_h - baseline), (x + text_w, y), (255,0,0), -1)
                    cv2.putText(image, text, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 2)
                    #cv2.rectangle(image, (x, y), (x + w, y + h), color, 2)  ##左上，右下
                    p = x + w / 2.
                    q = y + h / 2.
                    list_downoids = [(p, q)]
                    d = list_downoids
                    for f in d:
                        transformed_downoids.append(f)
                    corner_points_list = [(x, y), (x + w, y + h)]
                    for o in corner_points_list:
                        transformed_g_list.append(o)

        distance_minimum = 100
        if len(transformed_downoids) >= 2:
            # Iterate over every possible 2 by 2 between the points
            list_indexes = list(itertools.combinations(range(len(transformed_downoids)), 2))
            for i,pair in enumerate(itertools.combinations(transformed_downoids, r=2)):
                # if math.sqrt( (pair[0][0] - pair[1][0])**2 + (pair[0][1] - pair[1][1])**2 ) < int(distance_minimum):
                if math.sqrt(math.pow((pair[0][0] - pair[1][0]), 2) + math.pow((pair[0][1] - pair[1][1]), 2)) < int(distance_minimum) :
                    index_pt1 = list_indexes[i][0]
                    index_pt2 = list_indexes[i][1]
                    index_pt1_1 = index_pt1 * 2
                    index_pt2_1 = index_pt2 * 2
                    cv2.rectangle(image, transformed_g_list[index_pt1_1], transformed_g_list[index_pt1_1 + 1],(0, 0, 255), 2)
                    cv2.rectangle(image, transformed_g_list[index_pt2_1], transformed_g_list[index_pt2_1 + 1],(0, 0, 255), 2)

        out.write(image)
        cv2.imshow("Image", image)
        c = cv2.waitKey(33) & 0xff
        if c == 27:
            break
    capture.release()
    # out.release()
    cv2.destroyAllWindows()
video_demo()

4.视频

可参考The Multiple Object Tracking Benchmark!