YOLOv3 Object Detection 2

from google.colab import drive

drive.mount('/content/drive')

 

1. YOLOv3 Weight Download

1-1. 'yolov3.weights'

from urllib.request import urlretrieve

urlretrieve('https://pjreddie.com/media/files/yolov3.weights','yolov3.weights')

('yolov3.weights', <http.client.HTTPMessage at 0x7fe9928e0b50>)

---

1-2. 'yolov3-tiny.weights'

from urllib.request import urlretrieve

urlretrieve('https://pjreddie.com/media/files/yolov3-tiny.weights','yolov3-tiny.weights')

('yolov3-tiny.weights', <http.client.HTTPMessage at 0x7fe992250a10>)

---

2. darknet Clone

!git clone https://github.com/pjreddie/darknet.git

2-1. 'yolov3.cfg'

!cat darknet/cfg/yolov3.cfg

검출 하고자 하는 classes의 수에 따라 filters, classes의 값을 변경해야 한다.

---

2-2. 'yolov3-tiny.cfg'

!cat darknet/cfg/yolov3-tiny.cfg

---

2-3. 'coco.names'

!cat darknet/data/coco.names

• 80가지 객체

 

---

3. Image

3-1. Import Packages

import cv2
from google.colab.patches import cv2_imshow
import numpy as np

---

3-2. 'Detection' 최소 신뢰도(확률) 지정

min_confidence = 0.5

---

3-3. Load YOLOv3

net = cv2.dnn.readNet('yolov3.weights', 'darknet/cfg/yolov3.cfg')
# net = cv2.dnn.readNet('yolov3-tiny.weights', 'darknet/cfg/yolov3-tiny.cfg')

# Detection된 Object(Class) List 배열 정의
classes = []   

# 80개의 Object(class)를 구분할 수 있는 Object의 이름을 classes 배열에 넣어준다.
with open('darknet/data/coco.names', 'r') as f:
   classes = [line.strip() for line in f.readlines()]   

layer_names = net.getLayerNames()

output_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]

# Object 마다 컬러를 하나씩 다르게 지정
colors = np.random.uniform(0, 255, size=(len(classes), 3))

---

3-4. 원본 Image

img = cv2.imread('/content/drive/My Drive/Colab Notebooks/datasets/testImage.jpeg')
img = cv2.resize(img, None, fx = 0.8, fy = 0.8)

height, width, channels = img.shape

cv2_imshow(img)

---

3-5. Image Object Detection

blob = cv2.dnn.blobFromImage(img, 0.00392, (416, 416), (0, 0, 0), True, crop = False)

net.setInput(blob)

outs = net.forward(output_layers)

class_ids = []     # Detection Class_ID 저장
confidences = []   # Detection Class_신뢰도(확률) 저장
boxes = []         # Detection Boxing_정보 저장

for out in outs:
    for detection in out:
        scores = detection[5:]
        class_id = np.argmax(scores)    # Detected Class_ID
        confidence = scores[class_id]   # Detected Class_신뢰도(확률)

        if confidence > min_confidence:  
            # Object Detected
            center_x = int(detection[0] * width)
            center_y = int(detection[1] * height)
            w = int(detection[2] * width)
            h = int(detection[3] * height)

            # Rectangle Coordinates
            x = int(center_x - w / 2)
            y = int(center_y - h / 2)

            boxes.append([x, y, w, h])             # Boxing_정보 추가
            confidences.append(float(confidence))  # 신뢰도(확률) 추가
            class_ids.append(class_id)             # Class_ID 추가

indexes = cv2.dnn.NMSBoxes(boxes, confidences, min_confidence, 0.4)

font = cv2.FONT_HERSHEY_PLAIN

for i in range(len(boxes)):
    if i in indexes:    
        x, y, w, h = boxes[i]
        label = str(classes[class_ids[i]]) 
        print(i, label)
        color = colors[i]
        cv2.rectangle(img, (x, y), (x + w, y + h), color, 1)
        cv2.rectangle(img, (x, y - 20), (x + w, y), color, -1)
        cv2.putText(img, label, (x + 5, y - 5), font, 1, (255, 255, 255), 1)

cv2_imshow(img)

 

3 car 4 car 5 person 6 car 7 car 8 car 9 car 10 bicycle

---

4. Video

4-1. Import Packages

import cv2
import numpy as np
import time
import io
import base64
from IPython.display import HTML

---

4-2. 원본 Video

video = io.open('/content/drive/My Drive/Colab Notebooks/datasets/testVideo.mp4', 'r+b').read()

encoded = base64.b64encode(video)

HTML(data='''<video width="70%" controls>
                <source src="data:video/mp4;base64,{0}" type="video/mp4"/>
             </video>'''.format(encoded.decode('ascii')))

---

4-3. 탐색 정보

file_name = '/content/drive/My Drive/Colab Notebooks/datasets/testVideo.mp4'

min_confidence = 0.5   

output_name = 'testVideo_output.mp4' 

elapsed_time = 0

def detectAndDisplay(frame):
    start_time = time.time()
    img = cv2.resize(frame, None, fx = 0.8, fy = 0.8)
    height, width, channels = img.shape

    blob = cv2.dnn.blobFromImage(img, 0.00392, (416, 416), (0, 0, 0), True, crop=False)

    net.setInput(blob)
    outs = net.forward(output_layers)

    # Showing Informations
    class_ids = []      # Class_ID 저장
    confidences = []    # Class_신뢰도(확률) 저장
    boxes = []          # Boxing_정보 저장

    for out in outs:
        for detection in out:
            scores = detection[5:]
            class_id = np.argmax(scores)      # Detected Class_ID
            confidence = scores[class_id]     # Detected Class_신뢰도(확률)
            if confidence > min_confidence:
                
                # Object Detected
                center_x = int(detection[0] * width)
                center_y = int(detection[1] * height)
                w = int(detection[2] * width)
                h = int(detection[3] * height)

                # Rectangle Coordinates
                x = int(center_x - w / 2)
                y = int(center_y - h / 2)

                boxes.append([x, y, w, h])             # Boxing_정보 추가
                confidences.append(float(confidence))  # 신뢰도(확률) 추가
                class_ids.append(class_id)             # Class_ID 추가
    
    indexes = cv2.dnn.NMSBoxes(boxes, confidences, min_confidence, 0.4)    
    font = cv2.FONT_HERSHEY_PLAIN
    for i in range(len(boxes)):
        if i in indexes:    
            x, y, w, h = boxes[i]
            label = "{}: {:.2f}".format(classes[class_ids[i]], confidences[i]*100)    
            print(i, label)
            color = colors[1] ###
            cv2.rectangle(img, (x, y), (x + w, y + h), color, 1)
            cv2.rectangle(img, (x, y - 20), (x + w, y), color, -1)
            cv2.putText(img, label, (x + 5, y - 5), font, 1, (255, 255, 255), 1)
    process_time = time.time() - start_time
    global elapsed_time
    elapsed_time += process_time   
    print("=== A frame took {:.3f} seconds".format(process_time))

    # 결과 Video 저장
    global writer
    if writer is None and output_name is not None:
        fourcc = cv2.VideoWriter_fourcc(*"MJPG")
        writer = cv2.VideoWriter(output_name, fourcc, 30,
                (img.shape[1], img.shape[0]), True)

    if writer is not None:
        writer.write(img)

---

4-4. Load YOLOv3

net = cv2.dnn.readNet('yolov3.weights', 'darknet/cfg/yolov3.cfg')  

classes = []  

with open('darknet/data/coco.names', 'r') as f:
    classes = [line.strip() for line in f.readlines()]
layer_names = net.getLayerNames()
output_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]
# colors = np.random.uniform(0, 255, size=(len(classes), 3))  

cap = cv2.VideoCapture(file_name)
writer = None
if not cap.isOpened:
    print('--(!)Error opening video capture')
    exit(0)
while True:
    ret, frame = cap.read()
    if frame is None:
        
        cap.release()
        
        writer.release()
        print('--(!) No captured frame -- Break!')
        print("elapsed time {:.3f} seconds".format(elapsed_time))
        break
    detectAndDisplay(frame)

---