### ch05 程式碼下載 ###
$ cd ~/Lab
$ wget https://max543.com/debugger/class/python02/PiCamera/Lab/code/ch05.zip
$ unzip ch05.zip



### ex5-1 (5-1.py) ###

import cv2

rect = ((220, 20), (370, 190))			# 設定 ROI 的座標範圍
(left, top), (right, bottom) = rect

# 自訂一個取出子畫面的函數
def roiarea(frame):
    return frame[top:bottom, left:right]

# 自訂一個將 ROI 區域的資料貼回原本畫面的函數     
def replaceroi(frame, roi):
    frame[top:bottom, left:right] = roi
    return frame

cap = cv2.VideoCapture(0)
ratio = cap.get(cv2.CAP_PROP_FRAME_WIDTH) / cap.get(cv2.CAP_PROP_FRAME_HEIGHT)
WIDTH = 400
HEIGHT = int(WIDTH / ratio)

while True:
    ret, frame = cap.read()
    frame = cv2.resize(frame, (WIDTH, HEIGHT))
    frame = cv2.flip(frame, 1)

    # 取出子畫面
    roi = roiarea(frame)
    roi = cv2.cvtColor(roi, cv2.COLOR_BGR2HSV)

    # 將處理完的子畫面貼回到原本畫面中
    frame = replaceroi(frame, roi)

    # 在 ROI 範圍處畫個框
    cv2.rectangle(frame, rect[0], rect[1], (0,0,255), 2)

    cv2.imshow('frame', frame)
    
    if cv2.waitKey(1) == 27: 
        cv2.destroyAllWindows()
        break



### ex5-2 (5-2.py) ###

import cv2

cap = cv2.VideoCapture('vtest.avi')			# OpenCV 範例影片 vtest.avi
tracker = cv2.TrackerCSRT_create()
roi = None							# 宣告一個變數 roi 來儲存欲追蹤的座標位置
while True:
    ret, frame = cap.read()
 
    if roi is None:
        roi = cv2.selectROI('frame', frame)

        if roi != (0, 0, 0, 0):			# 若有框出一個區域，則將矩形區域傳給追蹤演算法的 init() 函數
            tracker.init(frame, roi)

    success, rect = tracker.update(frame)		# # 物體移動到新的區域，update() 計算新區域的座標，將新座標交給 rectangle() 畫出一個矩形區域。
    
    if success: 
        (x, y, w, h) = [int(i) for i in rect]
        cv2.rectangle(frame, (x,y), (x+w, y+h), (0, 255, 0), 2)

    cv2.imshow('frame', frame)

    if cv2.waitKey(66) == 27:
        cv2.destroyAllWindows()
        break



### ex5-3 ###

# YOLO 官方網址
https://pjreddie.com/darknet/yolo/

# YOLO 預訓練模型的權重檔
https://pjreddie.com/media/files/yolov3.weights

# OpenCV + YOLO 物體辨識的官方教學文件
https://opencv-tutorial.readthedocs.io/en/latest/yolo/yolo.html



### ex5-4 (5-3.py) ###

import cv2
import numpy as np

classes = open("yolo/coco.names").read().strip().split('\n')				# 載入分類名稱檔 coco.names
np.random.seed(42)											# 指定亂數種子
colors = np.random.randint(0, 255, size = (len(classes), 3), dtype = "uint8")		# 使用亂數產生不同分類色彩值的 Numpy 陣列

net = cv2.dnn.readNetFromDarknet("yolo/yolov3.cfg", "yolo/yolov3.weights")		# 載入 YOLO3 模型的設定檔 (Cfg File) 和權重檔 (Weight File)
net.setPreferableBackend(cv2.dnn.DNN_BACKEND_OPENCV)						# 指定優先選擇 OpenCV 後台

# 決定輸出層
ln = net.getLayerNames()
ln = [ln[i - 1] for i in net.getUnconnectedOutLayers()]

# post_process() 函數處理物體辨識結果
def post_process(img, outputs, conf):
    H, W = img.shape[:2]			# 取出圖片的高 (H) 與寬 (W)

    boxes = []					# 物體的方框座標 (空串列)
    confidences = []				# 信心指數值 (空串列)
    classIDs = []					# 分類編號 (空串列)
    
    # 一一取出辨識出的物體
    for output in outputs:			
        scores = output[5:]			# 取出分數
        classID = np.argmax(scores)		# 取出分類編號
        confidence = scores[classID]	# 取得信心指數值
        
        # 取得的信心指數值 confidence ，若大於所傳入的參數值 conf (傳入的 0.5)，則表示偵測到物體
        if confidence > conf:
            x, y, w, h = output[:4] * np.array([W, H, W, H])	# 取出物體範圍
            p0 = int(x - w//2), int(y - h//2)
            p1 = int(x + w//2), int(y + h//2)
            boxes.append([*p0, int(w), int(h)])
            confidences.append(float(confidence))
            classIDs.append(classID)            
    
    indices = cv2.dnn.NMSBoxes(boxes, confidences, conf, conf - 0.1)

    if len(indices) > 0:
        for i in indices.flatten():
            (x, y) = (boxes[i][0], boxes[i][1])
            (w, h) = (boxes[i][2], boxes[i][3])
            color = [int(c) for c in colors[classIDs[i]]]
            cv2.rectangle(img, (x, y), (x + w, y + h), color, 2)
            text = "{}: {:.4f}".format(classes[classIDs[i]], confidences[i])
            cv2.putText(img, text, (x, y - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 1)

img0 = cv2.imread("horse.jpg")
img = img0.copy()
blob = cv2.dnn.blobFromImage(img, 1/255.0, (416, 416), swapRB = True, crop = False)

net.setInput(blob)
outputs = net.forward(ln)
outputs = np.vstack(outputs)
post_process(img, outputs, 0.5)
cv2.imshow("window",  img)

cv2.waitKey(0)
cv2.destroyAllWindows()