{"id":38858,"date":"2024-11-02T01:39:37","date_gmt":"2024-11-02T01:39:37","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=38858"},"modified":"2024-11-26T06:43:10","modified_gmt":"2024-11-26T06:43:10","slug":"yolo-%eb%aa%a8%eb%8d%b8%ec%9d%84-%ec%9d%b4%ec%9a%a9%ed%95%9c-%ea%b0%9d%ec%b2%b4-%ec%b6%94%ec%a0%81%ea%b3%bc-%eb%8b%a4%ec%a4%91-%ea%b0%9d%ec%b2%b4-%ea%b2%80%ec%b6%9c-%ea%b0%9d%ec%b2%b4-%ea%b2%80","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/38858\/","title":{"rendered":"YOLO \ubaa8\ub378\uc744 \uc774\uc6a9\ud55c \uac1d\uccb4 \ucd94\uc801\uacfc \ub2e4\uc911 \uac1d\uccb4 \uac80\ucd9c, \uac1d\uccb4 \uac80\ucd9c\uacfc \uac1d\uccb4 \ucd94\uc801\uc758 \ucc28\uc774\uc810 \uc774\ud574\ud558\uae30"},"content":{"rendered":"

<\/p>\n

\ud604\ub300\uc758 \ucef4\ud4e8\ud130 \ube44\uc804 \ubd84\uc57c\uc5d0\uc11c, \uac1d\uccb4 \uac80\ucd9c(Object Detection)\uacfc \uac1d\uccb4 \ucd94\uc801(Object Tracking)\uc740 \ud544\uc218\uc801\uc778 \uae30\uc220\uc785\ub2c8\ub2e4. \uc774 \ub450 \uae30\uc220\uc740 \uc5ec\ub7ec \uc0b0\uc5c5\uacfc \uc751\uc6a9 \ud504\ub85c\uadf8\ub7a8\uc5d0\uc11c \ud65c\ubc1c\ud558\uac8c \uc0ac\uc6a9\ub418\uace0 \uc788\uc73c\uba70, \ud2b9\ud788 \uc790\uc728\uc8fc\ud589\ucc28\uc640 \uac10\uc2dc \uc2dc\uc2a4\ud15c \ub4f1\uc5d0\uc11c \uc911\uc694\ud569\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 YOLO(You Only Look Once) \ubaa8\ub378\uc744 \uae30\ubc18\uc73c\ub85c \uac1d\uccb4 \uac80\ucd9c\uacfc \uac1d\uccb4 \ucd94\uc801\uc758 \uac1c\ub150\uc744 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

1. YOLO \ubaa8\ub378 \uc18c\uac1c<\/h2>\n

YOLO\ub294 \uac1d\uccb4 \uac80\ucd9c\uc744 \uc704\ud55c \ub525\ub7ec\ub2dd \uae30\ubc18 \ubaa8\ub378\ub85c, 2015\ub144 Joseph Redmon\uc5d0 \uc758\ud574 \ucc98\uc74c \uc18c\uac1c\ub418\uc5c8\uc2b5\ub2c8\ub2e4. YOLO\uc758 \uac00\uc7a5 \ud070 \ud2b9\uc9d5\uc740 \uc774\ubbf8\uc9c0\ub97c \ud55c \ubc88\ub9cc \ubcf4\uace0 \uac1d\uccb4\ub97c \uac80\ucd9c\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \uc774\ub294 \uae30\uc874\uc758 \ubc29\ubc95\uc5d0 \ube44\ud574 \ud6e8\uc52c \ube60\ub978 \uc18d\ub3c4\ub85c \uac1d\uccb4\ub97c \uac80\ucd9c\ud560 \uc218 \uc788\uac8c \ud574\uc90d\ub2c8\ub2e4.<\/p>\n

1.1 YOLO\uc758 \uc791\ub3d9 \ubc29\uc2dd<\/h3>\n

YOLO\ub294 \uc785\ub825 \uc774\ubbf8\uc9c0\ub97c SxS \uadf8\ub9ac\ub4dc\ub85c \ub098\ub204\uace0, \uac01 \uadf8\ub9ac\ub4dc\uc5d0 \ub300\ud574 \ubc14\uc6b4\ub529 \ubc15\uc2a4\uc640 \ud074\ub798\uc2a4 \ud655\ub960\uc744 \uc608\uce21\ud569\ub2c8\ub2e4. \uac01 \uadf8\ub9ac\ub4dc\ub294 \ud2b9\uc815 \uac1d\uccb4\uac00 \uc874\uc7ac\ud560 \ud655\ub960\uc774 0.5 \uc774\uc0c1\uc77c \ub54c\ub9cc \uadf8 \uac1d\uccb4\ub97c \uac80\ucd9c\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc811\uadfc \ubc29\uc2dd\uc740 \uc18d\ub3c4\uc640 \uc131\ub2a5\uc744 \ub3d9\uc2dc\uc5d0 \uace0\ub824\ud55c \ub514\uc790\uc778\uc774\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n

2. \uac1d\uccb4 \uac80\ucd9c(Object Detection)\uacfc \uac1d\uccb4 \ucd94\uc801(Object Tracking)<\/h2>\n

\uac1d\uccb4 \uac80\ucd9c\uacfc \uac1d\uccb4 \ucd94\uc801\uc740 \ube44\uc2b7\ud574 \ubcf4\uc774\uc9c0\ub9cc, \uac1c\ub150\uc801\uc73c\ub85c \ud070 \ucc28\uc774\uac00 \uc788\uc2b5\ub2c8\ub2e4. \uac1d\uccb4 \uac80\ucd9c\uc740 \uc8fc\uc5b4\uc9c4 \uc774\ubbf8\uc9c0 \ub0b4\uc5d0\uc11c \uac1d\uccb4\ub97c \uc2dd\ubcc4\ud558\uc5ec \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub97c \uadf8\ub9ac\ub294 \uc791\uc5c5\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \ubc18\uba74 \uac1d\uccb4 \ucd94\uc801\uc740 \uc601\uc0c1 \uc18d\uc5d0\uc11c \uc2dc\uac04\uc5d0 \ub530\ub77c \uc6c0\uc9c1\uc774\ub294 \uac1d\uccb4\ub97c \uc9c0\uc18d\uc801\uc73c\ub85c \uc2dd\ubcc4\ud558\uace0 \ucd94\uc801\ud558\ub294 \uacfc\uc815\uc785\ub2c8\ub2e4.<\/p>\n

2.1 \uac1d\uccb4 \uac80\ucd9c\uc758 \uc608<\/h3>\n

\uac1d\uccb4 \uac80\ucd9c\uc758 \uc8fc\uc694 \uacfc\uc815\uc740 \uc774\ubbf8\uc9c0\ub97c \uc785\ub825\uc73c\ub85c \ubc1b\uc544 \uac1d\uccb4\uc758 \uc885\ub958\uc640 \uc704\uce58\ub97c \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub85c \ubc18\ud658\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c YOLO \ubaa8\ub378\uc774 \ub9e4\uc6b0 \uc720\uc6a9\ud558\uac8c \ud65c\uc6a9\ub429\ub2c8\ub2e4.<\/p>\n

2.2 \uac1d\uccb4 \ucd94\uc801\uc758 \uc608<\/h3>\n

\uac1d\uccb4 \ucd94\uc801\uc740 \uc5ec\ub7ec \ud504\ub808\uc784\uc5d0 \uac78\uccd0 \ub3d9\uc77c\ud55c \uac1d\uccb4\uac00 \uc9c0\uc18d\uc801\uc73c\ub85c \uc874\uc7ac\ud558\ub294\uc9c0\ub97c \uac80\uc0ac\ud558\ub294 \uac83\uc744 \ud3ec\ud568\ud569\ub2c8\ub2e4. \uc989, \ud2b9\uc815 \uac1d\uccb4\uac00 \ud55c \uc704\uce58\uc5d0\uc11c \ub2e4\ub978 \uc704\uce58\ub85c \uc774\ub3d9\ud558\ub294 \ub3d9\uc548 \uadf8 \uac1d\uccb4\ub97c \uc2dd\ubcc4\ud558\uc5ec \ucd94\uc801\ud574\uc57c \ud569\ub2c8\ub2e4.<\/p>\n

3. YOLO\ub97c \uc774\uc6a9\ud55c \uac1d\uccb4 \uac80\ucd9c\uacfc \ub2e4\uc911 \uac1d\uccb4 \uac80\ucd9c<\/h2>\n

YOLO\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac1d\uccb4 \uac80\ucd9c\uacfc \ub2e4\uc911 \uac1d\uccb4 \uac80\ucd9c\uc744 \uc218\ud589\ud558\ub294 \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c \ud30c\uc774\uc36c \ucf54\ub4dc \uc608\uc81c\uc5d0\uc11c\ub294 OpenCV\uc640 YOLOv3 \ubaa8\ub378\uc744 \uc774\uc6a9\ud558\uc5ec \uac1d\uccb4\ub97c \uac80\ucd9c\ud558\ub294 \ubc29\ubc95\uc744 \uc124\uba85\ud569\ub2c8\ub2e4.<\/p>\n

\nimport cv2\nimport numpy as np\n\n# YOLO \ubaa8\ub378 \ub85c\ub4dc\ndef load_yolo():\n    net = cv2.dnn.readNet(\"yolov3.weights\", \"yolov3.cfg\")\n    layer_names = net.getLayerNames()\n    output_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]\n    return net, output_layers\n\n# \uac1d\uccb4 \uac80\ucd9c \ud568\uc218\ndef detect_objects(img, net, output_layers):\n    height, width = img.shape[:2]\n    blob = cv2.dnn.blobFromImage(img, 0.00392, (416, 416), (0, 0, 0), True, crop=False)\n    net.setInput(blob)\n    outs = net.forward(output_layers)\n    \n    class_ids = []\n    confidences = []\n    boxes = []\n\n    for out in outs:\n        for detection in out:\n            scores = detection[5:]\n            class_id = np.argmax(scores)\n            confidence = scores[class_id]\n            if confidence > 0.5:  # \uc2e0\ub8b0\ub3c4 0.5 \uc774\uc0c1\n                center_x = int(detection[0] * width)\n                center_y = int(detection[1] * height)\n                w = int(detection[2] * width)\n                h = int(detection[3] * height)\n\n                # \ubc14\uc6b4\ub529 \ubc15\uc2a4 \uc88c\ud45c\n                x = int(center_x - w \/ 2)\n                y = int(center_y - h \/ 2)\n\n                boxes.append([x, y, w, h])\n                confidences.append(float(confidence))\n                class_ids.append(class_id)\n\n    indices = cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4)\n    for i in indices:\n        box = boxes[i]\n        x, y, w, h = box\n        label = str(classes[class_ids[i]])\n        confidence = confidences[i]\n        color = [int(c) for c in np.random.randint(0, 255, size=3)]\n        cv2.rectangle(img, (x, y), (x + w, y + h), color, 2)\n        cv2.putText(img, label + \" \" + str(round(confidence, 2)), (x, y + 30), cv2.FONT_HERSHEY_PLAIN, 3, color, 3)\n\n# \ud074\ub798\uc2a4 \ub85c\ub4dc\ndef load_classes():\n    with open(\"coco.names\", \"r\") as f:\n        return [line.strip() for line in f.readlines()]\n\n# \uba54\uc778 \ud568\uc218\nif __name__ == \"__main__\":\n    net, output_layers = load_yolo()\n    classes = load_classes()\n    img = cv2.imread(\"image.jpg\")  # \uc774\ubbf8\uc9c0 \uacbd\ub85c\n    detect_objects(img, net, output_layers)\n    cv2.imshow(\"Image\", img)\n    cv2.waitKey(0)\n    cv2.destroyAllWindows()\n<\/code><\/pre>\n
4. YOLO\ub97c \uc774\uc6a9\ud55c \uac1d\uccb4 \ucd94\uc801<\/h2>\n
\uac1d\uccb4 \uac80\ucd9c \ud6c4, \uac1d\uccb4 \ucd94\uc801\uc744 \uc9c4\ud589\ud558\uae30 \uc704\ud574 \uc5ec\ub7ec \uac00\uc9c0 \ubc29\ubc95\uc774 \uc788\uc9c0\ub9cc, \ub300\ud45c\uc801\uc73c\ub85c SORT(Simple Online and Realtime Tracking) \uc54c\uace0\ub9ac\uc998\uacfc Kalman \ud544\ud130\ub97c \ud65c\uc6a9\ud55c \ubc29\ubc95\uc774 \uc788\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c\uc740 YOLO\uc640 SORT\ub97c \uacb0\ud569\ud558\uc5ec \uac1d\uccb4 \ucd94\uc801\uc744 \uc218\ud589\ud558\ub294 \uc608\uc81c \ucf54\ub4dc\uc785\ub2c8\ub2e4.<\/p>\n
\nfrom sort import Sort\n\n# \uac1d\uccb4 \ucd94\uc801 \ucd08\uae30\ud654\ntracker = Sort()\n\ncap = cv2.VideoCapture(\"video.mp4\")  # \ube44\ub514\uc624 \uacbd\ub85c\nwhile cap.isOpened():\n    ret, frame = cap.read()\n    if not ret:\n        break\n    \n    detections = detect_objects(frame, net, output_layers)\n    detections = np.array(detections).reshape(-1, 5)  # [x, y, w, h, confidence]\n    \n    # SORT\ub97c \ud1b5\ud574 \uac1d\uccb4 \ucd94\uc801\n    trackers = tracker.update(detections)\n    \n    # \ucd94\uc801 \uacb0\uacfc \uc2dc\uac01\ud654\n    for d in trackers:\n        x1, y1, x2, y2, obj_id = d\n        cv2.rectangle(frame, (int(x1), int(y1)), (int(x2), int(y2)), (255, 0, 0), 2)\n        cv2.putText(frame, f'ID: {int(obj_id)}', (int(x1), int(y1)-10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)\n\n    cv2.imshow(\"Tracking\", frame)\n    if cv2.waitKey(1) & 0xFF == ord('q'):\n        break\n\ncap.release()\ncv2.destroyAllWindows()\n<\/code><\/pre>\n
5. \uac1d\uccb4 \uac80\ucd9c\uacfc \uac1d\uccb4 \ucd94\uc801\uc758 \ucc28\uc774\uc810<\/h2>\n
\uac1d\uccb4 \uac80\ucd9c\uc740 \uac1c\ubcc4 \ud504\ub808\uc784\uc5d0\uc11c \uac1d\uccb4\uc758 \uc704\uce58\uc640 \ud074\ub798\uc2a4\ub97c \uc2dd\ubcc4\ud558\ub294 \ubc18\uba74, \uac1d\uccb4 \ucd94\uc801\uc740 \uc5f0\uc18d\ub41c \ud504\ub808\uc784\uc5d0\uc11c \ud2b9\uc815 \uac1d\uccb4\ub97c \ucd94\uc801\ud558\uba70 \uadf8 \uc704\uce58\ub97c \uc5c5\ub370\uc774\ud2b8\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc989, \uac1d\uccb4 \uac80\ucd9c\uc740 \uc815\uc801\uc778 \uc791\uc5c5\uc774\uace0, \uac1d\uccb4 \ucd94\uc801\uc740 \ub3d9\uc801\uc778 \uc791\uc5c5\uc774\ub77c\uace0 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uac1d\uccb4 \uac80\ucd9c\uc744 \ud1b5\ud574 \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub97c \uc5bb\uc740 \ud6c4, \uac1d\uccb4 \ucd94\uc801\uc744 \ud1b5\ud574 \uadf8 \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub4e4\uc774 \uc5b4\ub5bb\uac8c \uc6c0\uc9c1\uc774\ub294\uc9c0\ub97c \uc9c0\uc18d\uc801\uc73c\ub85c \uc54c\uc544\ub0b4\ub294 \uac83\uc774 \uc774 \ub450 \uac1c\ub150\uc758 \uae30\ubcf8\uc801\uc778 \ucc28\ubcc4\uc810\uc785\ub2c8\ub2e4.<\/p>\n
5.1 \uc131\ub2a5 \ube44\uad50<\/h3>\n
YOLO\ub294 \ub9e4\uc6b0 \ube60\ub974\uace0 \ud6a8\uc728\uc801\uc778 \uac1d\uccb4 \uac80\ucd9c \ubc29\ubc95\uc73c\ub85c \ub110\ub9ac \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uadf8\ub7ec\ub098, \uac1d\uccb4 \ucd94\uc801\uc758 \uacbd\uc6b0, \ub9c8\uc9c0\ub9c9\uc73c\ub85c \uac80\ucd9c\ub41c \uc704\uce58\ub97c \uae30\ubc18\uc73c\ub85c \ub2e4\uc74c \uc704\uce58\ub97c \uc608\uce21\ud558\uae30 \ub54c\ubb38\uc5d0, \uac80\ucd9c\ub41c \uac1d\uccb4\uac00 \ud504\ub808\uc784 \ub0b4\uc5d0\uc11c \uc0ac\ub77c\uc9c0\uac70\ub098 \uacb9\uce58\ub294 \uacbd\uc6b0 \uc815\ud655\uc131\uc774 \ub0ae\uc544\uc9c8 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 Kalman \ud544\ud130\uc640 \uac19\uc740 \uc608\uce21 \uc54c\uace0\ub9ac\uc998\uc774 \uc0ac\uc6a9\ub429\ub2c8\ub2e4.<\/p>\n
5.2 \uc2e4\uc81c \uc801\uc6a9 \uc0ac\ub840<\/h3>\n
YOLO \ubaa8\ub378\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \uc0ac\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n