\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 OpenCV\ub97c \uc0ac\uc6a9\ud558\uc5ec YOLO(You Only Look Once)\uc640 SSD(Single Shot MultiBox Detector) \ubaa8\ub378\uc744 \ub85c\ub4dc\ud558\uace0 \ud65c\uc6a9\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774 \ub450 \uac00\uc9c0 \ubaa8\ub378\uc740 \uac1d\uccb4 \ud0d0\uc9c0 \ubd84\uc57c\uc5d0\uc11c \ub110\ub9ac \uc0ac\uc6a9\ub418\uba70, \uac01\uac01\uc758 \ud2b9\uc131\uacfc \uc7a5\ub2e8\uc810\uc774 \uc788\uc2b5\ub2c8\ub2e4. \ub610\ud55c, OpenCV\ub97c \uc0ac\uc6a9\ud55c \ub2e4\uc591\ud55c \uc774\ubbf8\uc9c0 \ubc0f \ube44\ub514\uc624 \ucc98\ub9ac \ubc29\ubc95\uc744 \ud568\uaed8 \uc18c\uac1c\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
OpenCV(Open Source Computer Vision Library)\ub294 \uc774\ubbf8\uc9c0 \ubc0f \ube44\ub514\uc624 \ucc98\ub9ac\ub97c \uc704\ud55c \uc624\ud508 \uc18c\uc2a4 \ub77c\uc774\ube0c\ub7ec\ub9ac\uc785\ub2c8\ub2e4. \uac15\ub825\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud558\uc5ec \ucef4\ud4e8\ud130 \ube44\uc804\uacfc \uba38\uc2e0 \ub7ec\ub2dd \ubd84\uc57c\uc5d0\uc11c \ub9ce\uc774 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \ub2e4\uc591\ud55c \ud504\ub85c\uadf8\ub798\ubc0d \uc5b8\uc5b4\uc5d0\uc11c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc73c\uba70, \ud2b9\ud788 Python\uacfc C++\uc5d0\uc11c \ub9ce\uc774 \ud65c\uc6a9\ub429\ub2c8\ub2e4. \uace0\uac1d \ub9de\ucda4\ud615 \uc5b4\ud50c\ub9ac\ucf00\uc774\uc158\uc744 \uc27d\uac8c \uac1c\ubc1c\ud560 \uc218 \uc788\ub3c4\ub85d \ub3c4\uc640\uc90d\ub2c8\ub2e4.<\/p>\n
YOLO(You Only Look Once)\ub294 \uac1d\uccb4 \ud0d0\uc9c0\ub97c \uc704\ud55c \ub525 \ub7ec\ub2dd \uae30\ubc18\uc758 \uc54c\uace0\ub9ac\uc998\uc785\ub2c8\ub2e4. YOLO\uc758 \uac00\uc7a5 \ud070 \uc7a5\uc810\uc740 \ub3d9\uc601\uc0c1\uc774\ub098 \uc2e4\uc2dc\uac04 \uc774\ubbf8\uc9c0\ub97c \ucc98\ub9ac\ud560 \ub54c \ub9e4\uc6b0 \ube60\ub978 \uc18d\ub3c4\ub85c \ub192\uc740 \uc815\ud655\ub3c4\uc758 \uac1d\uccb4 \ud0d0\uc9c0\ub97c \uac00\ub2a5\ud558\uac8c \ud55c\ub2e4\ub294 \uac83\uc785\ub2c8\ub2e4. YOLO\ub294 \uc774\ubbf8\uc9c0 \uc804\uccb4\ub97c \ud55c \ubc88\ub9cc \ubcf4\uace0, \uc5ec\ub7ec \uac1d\uccb4\uc758 \uc704\uce58\uc640 \uacc4\uce35\uc744 \ub3d9\uc2dc\uc5d0 \uc608\uce21\ud569\ub2c8\ub2e4.<\/p>\n
YOLO\ub294 \uc785\ub825 \uc774\ubbf8\uc9c0\ub97c SxS\uc758 \uadf8\ub9ac\ub4dc\ub85c \ub098\ub208 \ub2e4\uc74c \uac01 \uadf8\ub9ac\ub4dc\uac00 \ud0d0\uc9c0\ud558\ub294 \uac1d\uccb4\uc758 \ubc14\uc6b4\ub529 \ubc15\uc2a4\uc640 \ud074\ub798\uc2a4 \ud655\ub960\uc744 \uc608\uce21\ud558\ub294 \ubc29\uc2dd\uc73c\ub85c \ub3d9\uc791\ud569\ub2c8\ub2e4. \uac01 \uadf8\ub9ac\ub4dc\uc140\uc740 \uace0\uc720\ud55c \ubc14\uc6b4\ub529 \ubc15\uc2a4\uc640 \uac1d\uccb4 \ud074\ub798\uc2a4\ub97c \uc608\uce21\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uad6c\uc870\ub294 CNN(Convolutional Neural Network)\uc744 \uae30\ubc18\uc73c\ub85c \ud558\uba70, \uc2e0\uc18d\ud55c \ucc98\ub9ac \uc18d\ub3c4\ub97c \uc790\ub791\ud569\ub2c8\ub2e4.<\/p>\n
\ud604\uc7ac YOLO\ub294 \uc5ec\ub7ec \uac00\uc9c0 \ubc84\uc804\uc774 \uc874\uc7ac\ud569\ub2c8\ub2e4. \uadf8 \uc911\uc5d0\uc11c\ub3c4 YOLOv5\ub294 \uac00\uc7a5 \uc720\uc5f0\ud558\uace0 \ubc1c\uc804\ub41c \ud615\ud0dc\ub85c, PyTorch\ub85c \uc791\uc131\ub418\uc5b4 \uc788\uc5b4 \uc27d\uac8c \uc0ac\uc6a9\ud558\uace0 \uc2e4\ud5d8\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
SSD(Single Shot MultiBox Detector)\ub294 \ube60\ub978 \uc18d\ub3c4\uc640 \ub192\uc740 \uc815\ud655\ub3c4\ub97c \uac00\uc9c4 \ub610 \ub2e4\ub978 \uac1d\uccb4 \ud0d0\uc9c0 \uc54c\uace0\ub9ac\uc998\uc785\ub2c8\ub2e4. SSD\ub294 \ub2e4\uc591\ud55c \ud06c\uae30\uc758 \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub2e4\uc591\ud55c \ud06c\uae30\uc758 \uac1d\uccb4\ub97c \ud0d0\uc9c0\ud569\ub2c8\ub2e4. \uc774 \ubc29\uc2dd\uc740 \uc5ec\ub7ec \uac1c\uc758 \uc2a4\ucf00\uc77c\uc744 \uc801\uc6a9\ud558\uc5ec \uc9c4\ud589\ub418\uba70, \uc785\ub825 \uc774\ubbf8\uc9c0\ub97c \uc11c\ub85c \ub2e4\ub978 \uc2a4\ucf00\uc77c\uc758 \uba40\ud2f0 \ub808\uc774\ube14\ub85c \ucc98\ub9ac\ud569\ub2c8\ub2e4.<\/p>\n
SSD\ub294 CNN \uae30\ubc18\uc758 \uad6c\uc870\ub85c \ub2e4\uc591\ud55c \ud06c\uae30\uc758 \ud504\ub808\uc784\uc6cc\ud06c(Feature Map)\uc5d0\uc11c \uc5ec\ub7ec \ube44\uc728\uc758 \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub97c \uc0dd\uc131\ud558\uace0, \uac01 \ubc14\uc6b4\ub529 \ubc15\uc2a4\uc5d0 \ub300\ud574 \ud074\ub798\uc2a4\uc640 \ubc14\uc6b4\ub529 \ubc15\uc2a4\ub97c \uc608\uce21\ud569\ub2c8\ub2e4. \ub0b4\ubd80\uc5d0 \uc788\ub294 \uc5ec\ub7ec \uacc4\uce35\uc744 \ud1b5\ud574 \ub2e4\uc591\ud55c \ud06c\uae30\uc758 \uac1d\uccb4\ub97c \ud0d0\uc9c0\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
YOLO \ubaa8\ub378\uc744 \ub85c\ub4dc\ud558\uae30 \uc704\ud574 \uba3c\uc800 \ud544\uc694 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud558\uace0, \ubaa8\ub378\uacfc \ud074\ub798\uc2a4 \ud30c\uc77c\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud574\uc57c \ud569\ub2c8\ub2e4. \ub2e4\uc74c\uc740 YOLOv3 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uae30 \uc704\ud55c \ub2e8\uacc4\uc785\ub2c8\ub2e4.<\/p>\n
pip install opencv-python numpy<\/code><\/pre>\n4.1.2 YOLOv3 \ubaa8\ub378 \ud30c\uc77c \ub2e4\uc6b4\ub85c\ub4dc<\/h4>\n
github \ud639\uc740 YOLO \uacf5\uc2dd \uc0ac\uc774\ud2b8\uc5d0\uc11c \ub2e4\uc74c \ud30c\uc77c\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud569\ub2c8\ub2e4:<\/p>\n
\nyolov3.weights<\/code><\/li>\nyolov3.cfg<\/code><\/li>\ncoco.names<\/code> (\ud074\ub798\uc2a4 \uc774\ub984 \ud30c\uc77c)<\/li>\n<\/ul>\n4.1.3 YOLOv3 \ubaa8\ub378 \ub85c\ub4dc \ucf54\ub4dc<\/h4>\nimport cv2\nimport numpy as np\n\n# YOLOv3 \ubaa8\ub378\uacfc \ud074\ub798\uc2a4 \ud30c\uc77c\uc758 \uacbd\ub85c \uc124\uc815\nweights_path = \"yolov3.weights\"\nconfig_path = \"yolov3.cfg\"\nclass_names_path = \"coco.names\"\n\n# \ud074\ub798\uc2a4 \uc774\ub984 \ub85c\ub4dc\nwith open(class_names_path, 'r') as f:\n classes = f.read().strip().split('\\n')\n\n# YOLO \ubaa8\ub378 \ub85c\ub4dc\nnet = cv2.dnn.readNet(weights_path, config_path)<\/code><\/pre>\n4.1.4 \uc774\ubbf8\uc9c0\uc5d0\uc11c \uac1d\uccb4 \ud0d0\uc9c0\ud558\uae30<\/h4>\ndef detect_objects(image):\n height, width = image.shape[:2]\n \n # \uc774\ubbf8\uc9c0\ub97c YOLO\uc5d0 \uc801\ud569\ud55c \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658\n blob = cv2.dnn.blobFromImage(image, 0.00392, (416, 416), (0, 0, 0), True, crop=False)\n net.setInput(blob)\n\n # \ucd9c\ub825 \ub808\uc774\uc5b4 \uc774\ub984 \uac00\uc838\uc624\uae30\n layer_names = net.getLayerNames()\n output_layers = [layer_names[i[0] - 1] for i in net.getUnconnectedOutLayers()]\n\n # \ud0d0\uc9c0\ub41c \uac1d\uccb4\n detections = net.forward(output_layers)\n\n boxes, confidences, class_ids = [], [], []\n \n for detection in detections:\n for obj in detection:\n scores = obj[5:]\n class_id = np.argmax(scores)\n confidence = scores[class_id]\n if confidence > 0.5: # \uc2e0\ub8b0\ub3c4 \ud544\ud130\ub9c1\n center_x = int(obj[0] * width)\n center_y = int(obj[1] * height)\n w = int(obj[2] * width)\n h = int(obj[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 # \ube44\ucd5c\ub300 \uc5b5\uc81c\n indices = cv2.dnn.NMSBoxes(boxes, confidences, 0.5, 0.4)\n\n for i in indices:\n i = i[0]\n box = boxes[i]\n x, y, w, h = box\n label = str(classes[class_ids[i]])\n cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 2)\n cv2.putText(image, label, (x, y + 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)\n\n return image\n\n# \ud14c\uc2a4\ud2b8 \uc774\ubbf8\uc9c0\uc5d0\uc11c \uac1d\uccb4 \ud0d0\uc9c0\nimage = cv2.imread('test.jpg')\noutput_image = detect_objects(image)\ncv2.imshow('Image', output_image)\ncv2.waitKey(0)\ncv2.destroyAllWindows()<\/code><\/pre>\n4.2 SSD \ubaa8\ub378 \ub85c\ub4dc\ud558\uae30<\/h3>\n
SSD \ubaa8\ub378\ub3c4 \ub9c8\ucc2c\uac00\uc9c0\ub85c OpenCV\uc5d0\uc11c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. SSD \uc608\uc81c \ucf54\ub4dc\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4.<\/p>\n
4.2.1 SSD \ubaa8\ub378 \ud30c\uc77c \ub2e4\uc6b4\ub85c\ub4dc<\/h4>\n
\ub2e4\uc74c \ud30c\uc77c\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud569\ub2c8\ub2e4:<\/p>\n
\nSSD_Mobilenet_v2.caffemodel<\/code><\/li>\nSSD_Mobilenet_v2.prototxt<\/code><\/li>\n<\/ul>\n4.2.2 SSD \ubaa8\ub378 \ub85c\ub4dc \ucf54\ub4dc<\/h4>\n# SSD \ubaa8\ub378 \ub85c\ub4dc\nssd_net = cv2.dnn.readNetFromCaffe('SSD_Mobilenet_v2.prototxt', 'SSD_Mobilenet_v2.caffemodel')<\/code><\/pre>\n4.2.3 \uc774\ubbf8\uc9c0\uc5d0\uc11c \uac1d\uccb4 \ud0d0\uc9c0\ud558\uae30<\/h4>\ndef detect_objects_ssd(image):\n height, width = image.shape[:2]\n \n # \uc774\ubbf8\uc9c0\ub97c SSD\uc5d0 \uc801\ud569\ud55c \ud615\uc2dd\uc73c\ub85c \ubcc0\ud658\n blob = cv2.dnn.blobFromImage(image, 0.007843, (300, 300), 127.5)\n ssd_net.setInput(blob)\n\n # \ud0d0\uc9c0\ub41c \uac1d\uccb4\n detections = ssd_net.forward()\n\n for i in range(detections.shape[2]):\n confidence = detections[0, 0, i, 2]\n if confidence > 0.5: # \uc2e0\ub8b0\ub3c4 \ud544\ud130\ub9c1\n class_id = int(detections[0, 0, i, 1])\n box = detections[0, 0, i, 3:7] * np.array([width, height, width, height])\n (x, y, x1, y1) = box.astype(\"int\")\n label = f'{class_names[class_id]}: {confidence:.2f}'\n\n cv2.rectangle(image, (x, y), (x1, y1), (0, 255, 0), 2)\n cv2.putText(image, label, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)\n\n return image\n\n# \ud14c\uc2a4\ud2b8 \uc774\ubbf8\uc9c0\uc5d0\uc11c \uac1d\uccb4 \ud0d0\uc9c0\nimage = cv2.imread('test.jpg')\noutput_image = detect_objects_ssd(image)\ncv2.imshow('Image', output_image)\ncv2.waitKey(0)\ncv2.destroyAllWindows()<\/code><\/pre>\n5. \uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 OpenCV\ub97c \uc0ac\uc6a9\ud558\uc5ec YOLO \ubc0f SSD \ubaa8\ub378\uc744 \ub85c\ub4dc\ud558\uace0 \uac1d\uccb4 \ud0d0\uc9c0\ub97c \uc218\ud589\ud558\ub294 \ubc29\ubc95\uc744 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \ub450 \ubaa8\ub378\uc740 \uac01\uac01 \uc7a5\ub2e8\uc810\uc744 \uac00\uc9c0\uace0 \uc788\uc73c\uba70, \uc0ac\uc6a9\uc790\uc758 \ud544\uc694\uc5d0 \ub530\ub77c \uc120\ud0dd\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. YOLO\ub294 \uc18d\ub3c4\uac00 \ube60\ub974\uace0, SSD\ub294 \uc5ec\ub7ec \ud06c\uae30\uc640 \ube44\uc728\uc758 \uac1d\uccb4\ub97c \ud0d0\uc9c0\ud558\ub294 \ub370 \uc720\ub9ac\ud569\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 OpenCV\uc640 \uadf8\uc640 \uad00\ub828\ub41c \ub2e4\uc591\ud55c \uae30\uc220\uc744 \ud65c\uc6a9\ud558\uc5ec \ub354 \ub9ce\uc740 \ud504\ub85c\uc81d\ud2b8\ub97c \uc9c4\ud589\ud574 \ub098\uac00\uc2dc\uae38 \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n
6. \ucc38\uace0 \uc790\ub8cc<\/h2>\n