def get_average_image_2(df_image):
cnt = len(df_image.values.tolist())
img_exp = None
df2 = df_image.copy(deep=True)
df2.columns = ['fn', 'flg']
df2['del'] = 0
print(df_image.shape)
for i, dt in enumerate(df_image.values.tolist()):
path_img, flg = dt
path_img = path_img.replace(
'/media/user1/Data/data/20201116/',
'/home/user1/yu_develop/202011161_sample/image/')
if os.path.exists(path_img) == False:
df2.iloc[i, 2] = 1
continue
if i % 10 == 0:
print(i, i / cnt, path_img)
print(path_img)
if img_exp is None:
img, color_org = load_image_from_npy(path_img)
img_exp = np.expand_dims(img, axis=0)
if flg == 0 and i > 0:
img, color_org = load_image_from_npy(path_img)
img_exp_i = np.expand_dims(img, axis=0)
img_exp = np.concatenate([img_exp, img_exp_i], axis=0)
print("img_exp shape", img_exp.shape)
img_avg = np.mean(img_exp, axis=0)
df3 = df2[df2['del'] == 0]
df3 = df3.loc[:, ['fn', 'flg']]
return np.array(img_avg, dtype=np.uint8), df3
def get_human_lst():
path = '/media/user1/Data/data/20201116/'
patho = '/media/user1/Data/data/Data_out_20201116/'
path_category_index = 'dt/category_index.npy'
# "https://tfhub.dev/tensorflow/centernet/hourglass_1024x1024_kpts/1"
# "https://tfhub.dev/tensorflow/centernet/resnet50v1_fpn_512x512_kpts/1"
# category_index = np.load(path_category_index, allow_pickle='TRUE').item()
path_mod = "https://tfhub.dev/tensorflow/efficientdet/d4/1"
human_dt_lst = object_detection_from_npy_all_file(path, patho, path_mod,
path_category_index)
pd.DataFrame(human_dt_lst).to_csv(f'{patho}/human_detect.csv',
header=None,
index=None)
human_dt_lst = pd.read_csv(f'{patho}/human_detect.csv', header=None)
image_sum = None
for i, dt in enumerate(human_dt_lst.values.tolist()):
path_img, flg = dt
if i == 0:
img, color_org = load_image_from_npy(path_img)
image_sum = np.zeros(img.shape)
if flg == 0:
img, color_org = load_image_from_npy(path_img)
image_sum = image_sum + img
image_average = image_sum / (i + 1)
image_average = np.array(image_average, dtype=np.uint8)
image_average_fp = f'{patho}/image_average.npy'
np.save(image_average_fp, image_average)
image_average = np.load(image_average_fp)
for i, dt in enumerate(human_dt_lst.values.tolist()):
path_img, flg = dt
if flg == 1:
pathoful = path_img.replace(path.replace('/', '\\'), patho)
img, color_org = load_image_from_npy(path_img)
image_diff = img - image_average
image_diff_fp = pathoful.replace('image.npy',
'moving_detection_diff.png')
# image_diff_fp=f'{pathoful}/moving_detection_diff.png'
cv2.imwrite(image_diff_fp, image_diff)
gray = cv2.cvtColor(image_average, cv2.COLOR_BGR2GRAY)
avg = gray.copy().astype("float")
img_detect, img_detect_concat, avg = detect_frame(img, avg)
pil_img = Image.fromarray(img_detect.astype(np.uint8))
image_diff_fp = pathoful.replace('image.npy',
'moving_detection.png')
pil_img.save(image_diff_fp)
def get_average_image(df_image):
image_sum = None
cnt = len(df_image.values.tolist())
for i, dt in enumerate(df_image.values.tolist()):
path_img, flg = dt
if i % 100 == 0:
print(i, i / cnt, path_img)
if i == 0:
img, color_org = load_image_from_npy(path_img)
image_sum = np.zeros(img.shape)
if flg == 0:
img, color_org = load_image_from_npy(path_img)
image_sum = image_sum + img
image_average = image_sum / (i + 1)
image_average = np.array(image_average, dtype=np.uint8)
return image_average
def get_human_lst_each_diff():
path = '/media/user1/Data/data/20201116/'
patho = '/media/user1/Data/data/Data_out_20201116/'
camlst = ['cam0', 'cam1', 'cam2']
for cam in camlst:
df = pd.read_csv(f'{patho}/human_detect_{cam}.csv', header=None)
df_human = df[df[1] == 1]
image_average_fp = f'{patho}/image_average_{cam}.npy'
image_average = np.load(image_average_fp)
cnt = len(df_human.values.tolist())
for i, dt in enumerate(df_human.values.tolist()):
path_img, flg = dt
print(i, i / cnt, path_img)
pathoful = path_img.replace(path, patho)
img, color_org = load_image_from_npy(path_img)
image_diff = img - image_average
image_diff_fp = pathoful.replace('image.npy',
'moving_detection_diff.png')
#print(image_diff_fp)
cv2.imwrite(image_diff_fp, image_diff)
gray = cv2.cvtColor(image_average, cv2.COLOR_BGR2GRAY)
avg = gray.copy().astype("float")
img_detect, img_detect_concat, avg = detect_frame(img, avg)
pil_img = Image.fromarray(img_detect.astype(np.uint8))
image_diff_fp = pathoful.replace('image.npy',
'moving_detection.png')
#print(image_diff_fp)
pil_img.save(image_diff_fp)
def get_human_lst_each_diff_fun(path, patho, func=None, ext='cv2'):
# path = '/home/user1/yu_develop/202011161_sample/image/'
# patho = '/home/user1/yu_develop/202011161_sample/out/'
camlst = ['cam0', 'cam1', 'cam2']
if func is None:
return None, None
for cam in camlst:
df = pd.read_csv(f'{patho}/human_detect_{cam}_s.csv', header=None)
df_human = df[df[1] == 1]
image_average_fp = f'{patho}/image_average_{cam}_s.npy'
image_average = np.load(image_average_fp)
image_average = cv2.GaussianBlur(image_average, (5, 5), 0)
image_average_gray = cv2.cvtColor(image_average, cv2.COLOR_BGR2GRAY)
cnt = len(df_human.values.tolist())
for i, dt in enumerate(df_human.values.tolist()):
path_img, flg = dt
print(i, i / cnt, path_img)
path_img = path_img.replace(
'/media/user1/Data/data/20201116/',
'/home/user1/yu_develop/202011161_sample/image/')
pathoful = path_img.replace(path, patho)
os.makedirs(pathoful, exist_ok=True)
img, color_org = load_image_from_npy(path_img)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# image_diff, image_diff2 = get_diff_of_two_image_2(img_gray, image_average_gray)
# image1, image_diff2, image_diff = compute_difference_cv2(image_average_gray, img, color=[255, 0, 0])
image1, image_diff2, image_diff = func(image_average_gray,
img,
color=[255, 0, 0])
image_diff_fp = pathoful.replace(
'image.npy', f'{ext}_moving_detection_diff.png')
cv2.imwrite(image_diff_fp, image_diff)
image_diff_fp = pathoful.replace(
'image.npy', f'{ext}_moving_detection_diff_2.png')
cv2.imwrite(image_diff_fp, image_diff2)
gray = cv2.cvtColor(image_average, cv2.COLOR_BGR2GRAY)
avg = gray.copy().astype("float")
img_detect, img_detect_concat, avg = detect_frame(img, avg)
pil_img = Image.fromarray(img_detect.astype(np.uint8))
image_diff_fp = pathoful.replace('image.npy',
f'{ext}_moving_detection.png')
# print(image_diff_fp)
pil_img.save(image_diff_fp)
path = '/media/user1/Data/data/20201116/'
patho = '/media/user1/Data/data/Data_out_20201116/'
path_category_index = 'dt/category_index.npy'
# "https://tfhub.dev/tensorflow/centernet/hourglass_1024x1024_kpts/1"
# "https://tfhub.dev/tensorflow/centernet/resnet50v1_fpn_512x512_kpts/1"
# category_index = np.load(path_category_index, allow_pickle='TRUE').item()
path_mod = "https://tfhub.dev/tensorflow/efficientdet/d4/1"
human_dt_lst=object_detection_from_npy_all_file(path,patho,path_mod,path_category_index)
pd.DataFrame(human_dt_lst).to_csv(f'{patho}/human_detect.csv',header=None,index=None)
human_dt_lst=pd.read_csv(f'{patho}/human_detect.csv',header=None)
image_sum=None
for i,dt in enumerate(human_dt_lst.values.tolist()):
path_img,flg=dt
if i==0:
img, color_org = load_image_from_npy(path_img)
image_sum=np.zeros(img.shape)
if flg==1:
img, color_org = load_image_from_npy(path_img)
image_sum=image_sum+img
image_average=image_sum/(i+1)
image_average = np.array(image_average, dtype=np.uint8)
image_average_fp=f'{patho}/image_average.npy'
np.save(image_average_fp, image_average)
image_average = np.load(image_average_fp)
for i,dt in enumerate(human_dt_lst.values.tolist()):
path_img,flg=dt
if flg==1:
pathoful=path_img.replace(path.replace('/','\\'),patho)
img, color_org = load_image_from_npy(path_img)
