def anay_video():
in_path = filedialog.askopenfilename(initialdir="./Cameras/" +
camera_ip + "/Videos")
out_path = "./Cameras/" + camera_ip + "/Videos" + '/out_temp.mp4'
json_file_path = "./Cameras/" + camera_ip + "/" + camera_ip + "_metadata.json"
# password
with open(json_file_path, "r") as json_file:
json_data = json.load(json_file)
cap = cv2.VideoCapture(in_path)
flag, frame = cap.read()
cv2.imwrite('./Cameras/' + camera_ip + '/temp.jpg', frame)
ori_reso = (frame.shape[1], frame.shape[0])
x, _ = region_select(camera_ip)
os.remove('./Cameras/' + camera_ip + '/temp.jpg')
VA = VideoAnalysis(in_path,
out_path,
json_data['resolution'],
ori_reso,
30,
json_data['frame_rate'],
sorted_refPt=x)
VA.mask_video()
arr = VA.mask_rcnn_apply()
interested_frame = VA.find_interested_frames(arr, 2)
VA.clip_video(interested_frame)
def run(video_id, video):
'''
Main function for running all feature extraction steps.
TODO: move this to "__main__"
ARGS:
video_id: unique video identifier
video: path to video file (mp4 format)
RETURNS:
dataframe with frame-by-frame analysis
'''
## Go through every frame in the video and extract features
cap = cv2.VideoCapture(video)
frame = None
## Output dataframes
video_df = pd.DataFrame()
img_quality_df = pd.DataFrame()
num_frames = cap.get(cv2.cv.CV_CAP_PROP_FRAME_COUNT)
#num_frames = 3
debug = 0
if debug == 0:
i_frame = 0
while i_frame < num_frames:
## Progress logging
if i_frame % 100 == 0:
print i_frame
if frame is not None:
prev_frame = frame.copy()
else:
prev_frame = None
ret,frame = cap.read()
## If there are no more frames, break out of loop
if ret == False:
break
frame_series = pd.DataFrame(index=[0])
frame_series['video_id'] = video_id
frame_series['frame_number'] = i_frame
frame_series['time'] = cap.get(cv2.cv.CV_CAP_PROP_POS_MSEC) / 1000.
##=======================
## Image quality metrics
##=======================
## Blur
blur_df = SceneAnalysis.get_blur(frame)
## Color Spectrum
color_df = SceneAnalysis.get_hsv_hists(frame)
img_quality_df = img_quality_df.append(pd.concat([frame_series, blur_df, color_df], axis=1), ignore_index=True)
#img_quality_df = img_quality_df.append(frame_series)
##=======================
## Video/motion metrics
##=======================
## Optical Flow
if prev_frame is not None:
flow_df = VideoUtilities.optical_flow_on_frame(frame, prev_frame)
else:
flow_df = None
## Merge results into single dataframe
#video_df = video_df.append(pd.concat([frame_series, blur_df, color_df, flow_df], axis=1, ignore_index=True))
#video_df = video_df.append(frame_series)
#cat['video_id'] = video_id
#cat = pd.concat([blur_df, color_df, flow_df], axis=1)
if flow_df is not None:
flow_series = flow_df.copy()
##NOTE: the below code is duplicated above - bad style
flow_series['video_id'] = video_id
flow_series['frame_number'] = i_frame #cap.get(cv2.cv.CV_CAP_PROP_POS_FRAMES-1)
flow_series['time'] = cap.get(cv2.cv.CV_CAP_PROP_POS_MSEC) / 1000.
#cat = pd.concat([frame_series, cat], axis=1)
video_df = video_df.append(flow_series, ignore_index=True)
i_frame += 1
else:
img_quality_df = pd.read_pickle(video_id + '.img_quality.pkl')
## After completing the frame-by-frame analysis, run video metrics
## scene changes
scene_change_df = VideoAnalysis.detect_cut(video, img_quality_df['time'])
img_quality_df['is_scene_transition'] = scene_change_df['is_scene_transition'].copy()
## Pickle model and save it to S3 or local directory
img_quality_df.to_pickle(video_id + '.img_quality.pkl')
video_df.to_pickle(video_id + '.flow.pkl')
return img_quality_df, video_df
flow_series = flow_df.copy()
##NOTE: the below code is duplicated above - bad style
flow_series['video_id'] = video_id
flow_series['frame_number'] = i_frame #cap.get(cv2.cv.CV_CAP_PROP_POS_FRAMES-1)
flow_series['time'] = cap.get(cv2.cv.CV_CAP_PROP_POS_MSEC) / 1000.
#cat = pd.concat([frame_series, cat], axis=1)
video_df = video_df.append(flow_series, ignore_index=True)
i_frame += 1
else:
img_quality_df = pd.read_pickle(video_id + '.img_quality.pkl')
## After completing the frame-by-frame analysis, run video metrics
## scene changes
scene_change_df = VideoAnalysis.detect_cut(video, img_quality_df['time'])
img_quality_df['is_scene_transition'] = scene_change_df['is_scene_transition'].copy()
## Pickle model and save it to S3 or local directory
img_quality_df.to_pickle(video_id + '.img_quality.pkl')
video_df.to_pickle(video_id + '.flow.pkl')
return img_quality_df, video_df
if __name__ == "__main__":
video = "../media/CKeLfaOl0Qk.mp4"
video_df = pd.read_pickle('CKeLfaOl0Qk.img_quality.pkl')
video_series = video_df['time']
scene_change_df = VideoAnalysis.detect_cut(video, video_series)
#print scene_change_df[scene_change_df['is_scene_transition'] == 1]
video_df['is_scene_transition'] = scene_change_df['is_scene_transition'].copy()
print video_df[video_df['is_scene_transition'] == 1]
def run(video_id, video):
'''
Main function for running all feature extraction steps.
TODO: move this to "__main__"
ARGS:
video_id: unique video identifier
video: path to video file (mp4 format)
RETURNS:
dataframe with frame-by-frame analysis
'''
## Go through every frame in the video and extract features
cap = cv2.VideoCapture(video)
frame = None
## Output dataframes
video_df = pd.DataFrame()
img_quality_df = pd.DataFrame()
num_frames = cap.get(cv2.cv.CV_CAP_PROP_FRAME_COUNT)
#num_frames = 3
debug = 0
if debug == 0:
i_frame = 0
while i_frame < num_frames:
## Progress logging
if i_frame % 100 == 0:
print i_frame
if frame is not None:
prev_frame = frame.copy()
else:
prev_frame = None
ret, frame = cap.read()
## If there are no more frames, break out of loop
if ret == False:
break
frame_series = pd.DataFrame(index=[0])
frame_series['video_id'] = video_id
frame_series['frame_number'] = i_frame
frame_series['time'] = cap.get(
cv2.cv.CV_CAP_PROP_POS_MSEC) / 1000.
##=======================
## Image quality metrics
##=======================
## Blur
blur_df = SceneAnalysis.get_blur(frame)
## Color Spectrum
color_df = SceneAnalysis.get_hsv_hists(frame)
img_quality_df = img_quality_df.append(pd.concat(
[frame_series, blur_df, color_df], axis=1),
ignore_index=True)
#img_quality_df = img_quality_df.append(frame_series)
##=======================
## Video/motion metrics
##=======================
## Optical Flow
if prev_frame is not None:
flow_df = VideoUtilities.optical_flow_on_frame(
frame, prev_frame)
else:
flow_df = None
## Merge results into single dataframe
#video_df = video_df.append(pd.concat([frame_series, blur_df, color_df, flow_df], axis=1, ignore_index=True))
#video_df = video_df.append(frame_series)
#cat['video_id'] = video_id
#cat = pd.concat([blur_df, color_df, flow_df], axis=1)
if flow_df is not None:
flow_series = flow_df.copy()
##NOTE: the below code is duplicated above - bad style
flow_series['video_id'] = video_id
flow_series[
'frame_number'] = i_frame #cap.get(cv2.cv.CV_CAP_PROP_POS_FRAMES-1)
flow_series['time'] = cap.get(
cv2.cv.CV_CAP_PROP_POS_MSEC) / 1000.
#cat = pd.concat([frame_series, cat], axis=1)
video_df = video_df.append(flow_series, ignore_index=True)
i_frame += 1
else:
img_quality_df = pd.read_pickle(video_id + '.img_quality.pkl')
## After completing the frame-by-frame analysis, run video metrics
## scene changes
scene_change_df = VideoAnalysis.detect_cut(video,
img_quality_df['time'])
img_quality_df['is_scene_transition'] = scene_change_df[
'is_scene_transition'].copy()
## Pickle model and save it to S3 or local directory
img_quality_df.to_pickle(video_id + '.img_quality.pkl')
video_df.to_pickle(video_id + '.flow.pkl')
return img_quality_df, video_df
cv2.cv.CV_CAP_PROP_POS_MSEC) / 1000.
#cat = pd.concat([frame_series, cat], axis=1)
video_df = video_df.append(flow_series, ignore_index=True)
i_frame += 1
else:
img_quality_df = pd.read_pickle(video_id + '.img_quality.pkl')
## After completing the frame-by-frame analysis, run video metrics
## scene changes
scene_change_df = VideoAnalysis.detect_cut(video,
img_quality_df['time'])
img_quality_df['is_scene_transition'] = scene_change_df[
'is_scene_transition'].copy()
## Pickle model and save it to S3 or local directory
img_quality_df.to_pickle(video_id + '.img_quality.pkl')
video_df.to_pickle(video_id + '.flow.pkl')
return img_quality_df, video_df
if __name__ == "__main__":
video = "../media/CKeLfaOl0Qk.mp4"
video_df = pd.read_pickle('CKeLfaOl0Qk.img_quality.pkl')
video_series = video_df['time']
scene_change_df = VideoAnalysis.detect_cut(video, video_series)
#print scene_change_df[scene_change_df['is_scene_transition'] == 1]
video_df['is_scene_transition'] = scene_change_df[
'is_scene_transition'].copy()
print video_df[video_df['is_scene_transition'] == 1]
def compute_analysis(records):
if (len(records)) == 0:
return
video_analysis = VideoAnalysis(records)
st.markdown(get_table_download_link(records), unsafe_allow_html=True)
st.header('Registros de detecciones')
st.dataframe(records)
df = video_analysis.get_complete_class_counts(as_seconds=True)
fig = px.bar(df, x='timestamp', y=df.columns, title='Cantidad de instancias detectadas durante el video')
fig.update_layout(
yaxis=dict(
title=dict(text='instancias'),
),
xaxis=dict(
title=dict(text='segundos'),
rangeslider=dict(
visible=True
),
type="linear"
)
)
st.plotly_chart(
fig)
st.header('Buscar rangos de tiempo donde aparecen un conjunto de clases')
single_class_time_tolerance = st.number_input('Tolerancia de separación de frames (en ms)', min_value=0, step=1000, value=1000)
s1_col1, s1_col2 = st.beta_columns(2)
with s1_col1:
class_selection = st.multiselect('Selecciona las clases que deseas buscar', options=video_analysis.get_unique_classes())
with s1_col2:
ranges = []
if len(class_selection) > 0:
ranges = video_analysis.get_timeranges_with_classes(class_selection, single_class_time_tolerance)
range_table = []
for rng in ranges:
range_table.append({
'Desde': rng[0] / 1000,
'Hasta': rng[1] / 1000
})
st.write('Rangos de tiempo donde aparecen las clases seleccionadas (en segundos)')
st.table(range_table)
instance_counts_selection = {}
st.header('Buscar rangos de tiempo donde aparecen una cantidad especifica de instancias')
instance_count_time_tolerance = st.number_input('Tolerancia de separación de frames (en ms)', min_value=0, step=1000, value=1000, key='instance_tolerance')
s2_col1, s2_col2 = st.beta_columns(2)
with s2_col1:
for classname in video_analysis.get_unique_classes():
instance_counts_selection[classname] = st.number_input(classname, min_value=0, value=0, step=1)
with s2_col2:
clean_selections = {}
for key in instance_counts_selection:
if instance_counts_selection[key] > 0:
clean_selections[key] = int(instance_counts_selection[key])
ranges = []
if len(clean_selections.keys()) > 0:
ranges = video_analysis.get_timeranges_by_instance_counts(clean_selections, instance_count_time_tolerance)
range_table = []
for rng in ranges:
range_table.append({
'Desde': rng[0] / 1000,
'Hasta': rng[1] / 1000
})
st.write('Rangos de tiempo donde aparecen las clases seleccionadas (en segundos)')
st.table(range_table)
def jpeg():
return Response(VideoAnalysis(**parameters).request_image(),
mimetype='image/jpeg',
direct_passthrough=True)
def mjpeg():
"""Video streaming route. Put this in the src attribute of an img tag."""
return Response(VideoAnalysis(**parameters).mjpeg_generator(),
mimetype='multipart/x-mixed-replace; boundary=frame',
direct_passthrough=True)
from region_select import region_select from tkinter import filedialog import json camera_ip = '[email protected]' root_path = '/home/ke/TrafficProject/Mask-RCNN/' in_path = filedialog.askopenfilename(initialdir=root_path + "/Cameras/" + camera_ip + "/Videos") out_path = '/home/ke/TrafficProject/Mask-RCNN/' + "/Cameras/" + camera_ip + "/Videos" + '/out_temp.mp4' json_file_path = root_path + "/Cameras/" + camera_ip + "/" + camera_ip + "_metadata.json" # password with open(json_file_path, "r") as json_file: json_data = json.load(json_file) cap = cv2.VideoCapture(in_path) flag, frame = cap.read() cv2.imwrite('./Cameras/haha/haha.jpg', frame) ori_reso = (frame.shape[1], frame.shape[0]) x, _ = region_select('haha') # A = VideoAnalysis('/home/ke/Downloads/test2.mp4', '/home/ke/Downloads/13.mp4', (640, 400), ori_reso, 30, 1) VA = VideoAnalysis(in_path, out_path, json_data['resolution'], ori_reso, 30, json_data['frame_rate'], sorted_refPt=x) VA.mask_video() arr = VA.mask_rcnn_apply() interested_frame = VA.find_interested_frames(arr, 2) VA.clip_video(interested_frame)
