def test_PF_ellipse_3(self):
test_path = "input_images/input_test/ellipse/3/"
template_path = "input_images/input_test/ellipse/template.png"
scene = self.get_scene_info(test_path, template_path)
frame, template_rect, template, img_path, img_list, points_array = scene
pf = ps5.AppearanceModelPF(frame, template,
num_particles=400,
sigma_exp=10., sigma_dyn=20.,
alpha=.05, template_coords=template_rect)
axes_shape = (50, 25)
self.run_filter(pf, img_path, img_list, points_array, axes_shape)
def part_5():
"""Tracking multiple Targets.
Use either a Kalman or particle filter to track multiple targets
as they move through the given video. Use the sequence of images
in the TUD-Campus directory.
Follow the instructions in the problem set instructions.
Place all your work in this file and this section.
"""
save_frames = {29: os.path.join(output_dir, 'ps5-5-a-1.png'),
56: os.path.join(output_dir, 'ps5-5-a-2.png'),
71: os.path.join(output_dir, 'ps5-5-a-3.png')}
t1 = {
'x': 60,
'y': 200,
'w': 100,
'h': 100
}
t2 = {
'x': 414,
'y': 220,
'w': 100,
'h': 100
}
t3 = {
'x': 20,
'y': 172,
'w': 60,
'h': 150
}
kwargs1 = {
'num_particles': 400,
'sigma_exp': 5,
'sigma_dyn': 15,
'alpha': 0.05,
}
kwargs2 = {
'num_particles': 250,
'sigma_exp': 5,
'sigma_dyn': 10,
'alpha': 0.,
}
kwargs3 = {
'num_particles': 150,
'sigma_exp': 5,
'sigma_dyn': 15,
'alpha': 0.05,
}
imgs_dir = os.path.join(input_dir, "TUD-Campus")
imgs_list = [f for f in os.listdir(imgs_dir)
if f[0] != '.' and f.endswith('.jpg')]
imgs_list = sorted(imgs_list)
# Initialize objects
templates = []
pf1 = None
pf2 = None
pf3 = None
frame_num = 1
for img in imgs_list:
frame = cv2.imread(os.path.join(os.path.join(input_dir, "TUD-Campus"), img))
# Extract template and initialize (one-time only)
if len(templates) < 1:
template1 = frame[int(t1['y']): int(t1['y'] + t1['h']), int(t1['x']): int(t1['x'] + t1['w'])]
template2 = frame[int(t2['y']):int(t2['y'] + t2['h']), int(t2['x']): int(t2['x'] + t2['w'])]
templates.append(template1)
templates.append(template2)
pf1 = ps5.AppearanceModelPF(frame, template=template1, template_coords=t1, **kwargs1)
pf2 = ps5.AppearanceModelPF(frame, template=template2, template_coords=t2, **kwargs2)
if frame_num == 32:
template3 = frame[int(t3['y']):int(t3['y'] + t3['h']), int(t3['x']): int(t3['x'] + t3['w'])]
templates.append(template3)
pf3 = ps5.AppearanceModelPF(frame, template=template3, template_coords=t3, **kwargs3)
# Process frame
pf1.process(frame)
if frame_num <= 29:
pf2.process(frame)
if frame_num >= 32:
pf3.process(frame)
if True: # For debugging, it displays every frame
out_frame = frame.copy()
pf1.render(out_frame)
if frame_num <= 29:
pf2.render(out_frame)
if frame_num >= 32:
pf3.render(out_frame)
cv2.imshow('Tracking', out_frame)
cv2.waitKey(1)
# Render and save output, if indicated
if frame_num in save_frames:
frame_out = frame.copy()
pf1.render(frame_out)
if frame_num <= 29:
pf2.render(frame_out)
if frame_num >= 32:
pf3.render(frame_out)
cv2.imwrite(save_frames[frame_num], frame_out)
# if frame_num == 71:
# frame_out = frame.copy()
# pf1.render(frame_out)
# pf2.render(frame_out)
# pf3.render(frame_out)
# cv2.imwrite(save_frames[frame_num], frame_out)
# Update frame number
frame_num += 1
def part_5():
"""Tracking multiple Targets.
Use either a Kalman or particle filter to track multiple targets
as they move through the given video. Use the sequence of images
in the TUD-Campus directory.
Follow the instructions in the problem set instructions.
Place all your work in this file and this section.
"""
save_frames = {29: os.path.join(output_dir, 'ps5-5-a-1.png'),
56: os.path.join(output_dir, 'ps5-5-a-2.png'),
71: os.path.join(output_dir, 'ps5-5-a-3.png')}
# kf = ps5.KalmanFilter(105, 260, R=0.2 * np.eye(2))
# template_rect = {'x': 55, 'y': 155, 'w': 100, 'h': 215}
#
# run_kalman_filter(kf,
# os.path.join(input_dir, "TUD-CAMPUS"),
# NOISE_1,
# "matching",
# save_frames,
# template_rect)
# kf = ps5.KalmanFilter(10, 235, R=0.2 * np.eye(2))
#
# kalman.run_kalman_filter(kf,
# os.path.join(input_dir, "TUD-CAMPUS"),
# NOISE_1,
# "matching",
# save_frames,
# None)
# kf = ps5.KalmanFilter(300, 235, R=0.2 * np.eye(2))
#
# kalman_2.run_kalman_filter(kf,
# os.path.join(input_dir, "TUD-CAMPUS"),
# NOISE_1,
# "matching",
# save_frames,
# None)
# ------------------------------------ 15, 26
num_particles_1 = 1300
sigma_md_1 = 10
sigma_dyn_1 = 18
filter_1 = None
box_1 = {'x_min': 290, 'x_max': 340, 'y_min': 175, 'y_max': 300}
num_particles_2 = 600
sigma_md_2 = 10
sigma_dyn_2 = 13
filter_2 = None
box_2 = {'x_min': 65, 'x_max': 175, 'y_min': 125, 'y_max': 355}
num_particles_3 = 800
sigma_md_3 = 30
sigma_dyn_3 = 15
filter_3 = None
box_3 = {'x_min': 0, 'x_max': 100, 'y_min': 125, 'y_max': 400}
imgs_list = [f for f in os.listdir("input_images/TUD-Campus")
if f[0] != '.' and f.endswith('.jpg')]
imgs_list.sort()
frame_num = 1
template_1 = get_template_1()
template_2 = get_template_2()
template_3 = get_template_3()
# Loop over video (till last frame or Ctrl+C is presssed)
for img in imgs_list:
frame = cv2.imread(os.path.join("input_images/TUD-Campus", img))
if filter_1 is None:
filter_1 = ps5.AppearanceModelPF(frame, template_1, num_particles=num_particles_1, sigma_exp=sigma_md_1,
sigma_dyn=sigma_dyn_1, alpha=0.0, in_gray_mode=False,
use_threshold=False, threshold=40.0,
use_box_initialization=True, box=box_1)
if filter_2 is None:
filter_2 = ps5.AppearanceModelPF(frame, template_2, num_particles=num_particles_2, sigma_exp=sigma_md_2,
sigma_dyn=sigma_dyn_2, alpha=0.0, in_gray_mode=False,
use_box_initialization=True, box=box_2)
if filter_3 is None:
filter_3 = ps5.AppearanceModelPF(frame, template_3, num_particles=num_particles_3, sigma_exp=sigma_md_3,
sigma_dyn=sigma_dyn_3, alpha=0.0, in_gray_mode=False,
use_box_initialization=True, box=box_3)
process_filters(filter_1, filter_2, filter_3, frame, frame_num, save_frames=save_frames)
frame_num += 1
