def detect_from_video():
detector = CustomVideoObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath(detection_model_path=os.path.join(execution_path, "detection_model-ex-33--loss-4.97.h5"))
detector.setJsonPath(configuration_json=os.path.join(execution_path, "detection_config.json"))
detector.loadModel()
detected_video_path = detector.detectObjectsFromVideo(input_file_path=os.path.join(execution_path, "video1.mp4"), frames_per_second=30, output_file_path=os.path.join(execution_path, "video1-detected"), minimum_percentage_probability=40, log_progress=True )
def detect_from_video(video_input):
detector = CustomVideoObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath(
detection_model_path=os.path.join(execution_path, model_path))
detector.setJsonPath(
configuration_json=os.path.join(execution_path, model_config))
detector.loadModel()
detected_video_path = detector.detectObjectsFromVideo(
input_file_path=os.path.join(execution_path, video_input),
per_frame_function=forFrame,
save_detected_video=False,
minimum_percentage_probability=40,
log_progress=True,
return_detected_frame=True)
def test_custom_video_detection_yolov3():
detector = CustomVideoObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath(model_path)
detector.setJsonPath(model_json)
detector.loadModel()
video_path = detector.detectObjectsFromVideo(
input_file_path=video_file,
output_file_path=video_file_output,
save_detected_video=True,
frames_per_second=30,
log_progress=True)
assert os.path.exists(video_file_output + ".avi")
assert isinstance(video_path, str)
os.remove(video_file_output + ".avi")
def recognize_custom(image_path, vehicle_type, isVideo):
if vehicle_type == 'TANK':
model_name = 'tank_model_3_loss-0008.h5'
json_name = 'tank_detection_config.json'
elif vehicle_type == 'APC':
model_name = 'apc_model_6_loss-0012.h5'
json_name = 'apc_detection_config.json'
elif vehicle_type == 'SHIP':
print('3')
elif vehicle_type == 'AERO':
print('4')
folder_path = str(
os.getcwd()) + "\\recognize_app\\rec_api\\imageairecognizer\\"
if isVideo == False:
detector = CustomObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath(folder_path + model_name)
detector.setJsonPath(folder_path + json_name)
detector.loadModel()
detector = detector.detectObjectsFromImage(
input_image=image_path,
output_image_path=image_path,
minimum_percentage_probability=30)
else:
video_detector = CustomVideoObjectDetection()
video_detector.setModelTypeAsYOLOv3()
video_detector.setModelPath(folder_path + model_name)
video_detector.setJsonPath(folder_path + json_name)
video_detector.loadModel()
print(Path(image_path).stem)
video_detector = video_detector.detectObjectsFromVideo(
input_file_path=image_path,
output_file_path=image_path,
minimum_percentage_probability=30,
frames_per_second=30,
log_progress=True)
os.remove(image_path)
p = Path(video_detector)
FILE_NAME = p.with_suffix('').with_suffix('').name
os.popen(
"ffmpeg -i {input} -ac 2 -b:v 2000k -c:a aac -c:v libx264 -b:a 160k -vprofile high -bf 0 -strict experimental -f mp4 {output}.mp4"
.format(input=p,
output='.\media\{0}'.format(FILE_NAME))).close()
def videoLoop():
execution_path = os.getcwd()
camera = cv2.VideoCapture(0)
video_detector = CustomVideoObjectDetection()
video_detector.setModelTypeAsYOLOv3()
video_detector.setModelPath("detection_model-ex-012--loss-0003.960.h5") #Download the model from "https://github.com/NAERSTEAM/EELAB/releases/download/20202603_0/detection_model-ex-012--loss-0003.960.h5"
video_detector.setJsonPath("detection_config HandSign.json")
video_detector.loadModel(detection_speed='fastest')
video_detector.detectObjectsFromVideo(camera_input=camera,
output_file_path=os.path.join(execution_path, "HandSignRecognition"),
frames_per_second=20,
minimum_percentage_probability=80,
log_progress=False,
save_detected_video=False,
per_frame_function=per_frame_function_DataGet,
return_detected_frame=True)
def videoLoop():
execution_path = os.getcwd()
camera = cv2.VideoCapture(0)
video_detector = CustomVideoObjectDetection()
video_detector.setModelTypeAsYOLOv3()
video_detector.setModelPath(
"detection_model-ex-012--loss-0003.944.h5"
) #Download the model from "https://github.com/firmamentone/masksDetection/releases/download/20200501_0/detection_model-ex-012--loss-0003.944.h5"
video_detector.setJsonPath("detection_config_Mask.json")
video_detector.loadModel(detection_speed='fastest')
video_detector.detectObjectsFromVideo(
camera_input=camera,
output_file_path=os.path.join(execution_path, "holo1-detected3"),
frames_per_second=20,
minimum_percentage_probability=40,
log_progress=True,
save_detected_video=False,
live_window=True) #AIT parameter:live_window
def detect_from_camera():
camera = cv2.VideoCapture(0)
# scale down video for better performance
camera.set(3, 320) # camera width
camera.set(4, 240) # camera height
camera.set(30, 0.1) #camera fps
detector = CustomVideoObjectDetection()
detector.setModelTypeAsYOLOv3()
detector.setModelPath(
detection_model_path=os.path.join(execution_path, model_path))
detector.setJsonPath(
configuration_json=os.path.join(execution_path, model_config))
detector.loadModel()
detected_video_path = detector.detectObjectsFromVideo(
camera_input=camera,
per_frame_function=forFrame,
save_detected_video=False,
minimum_percentage_probability=40,
log_progress=True,
return_detected_frame=True)
# Importing the image AI library
# this library is an API used object detection system
from imageai.Detection.Custom import CustomVideoObjectDetection
import os
import cv2
execution_path = os.getcwd()
# using openCV to start the camera for taking the video feed
camera = cv2.VideoCapture(0)
detector = CustomVideoObjectDetection()
# our model is trained using transfer learning on YOLOV3.
detector.setModelTypeAsYOLOv3()
# put your trained keras model path in the space provided.
detector.setModelPath("Path to trained keras model")
# put the configuration file's path over here
detector.setJsonPath("path to json configuration file")
detector.loadModel()
#here the video is taken from the camera and in the outpat path, put your path where you can access the output video feed.
#minimum_percentage_probability : you can increase or decrease to fine tune your output.
detector.detectObjectsFromVideo(camera_input=camera,
output_file_path=os.path.join(
execution_path, "name of output video"),
frames_per_second=16,
minimum_percentage_probability=40,
log_progress=True)