def test_KF_Color(self):
file_path = "input_images/input_test/color"
template = cv2.imread("input_images/input_test/template.jpg")
images, centers = load_data(file_path)
kf = ps5.KalmanFilter(*centers[0, :])
assert_scene(kf, images, centers, NOISE_0, template,
lambda x: self.assertAlmostEqual(x, 0, delta=15))
def part_1b():
print("Part 1b")
template_loc = {'y': 72, 'x': 140, 'w': 50, 'h': 50}
# Define process and measurement arrays if you want to use other than the
# default. Pass them to KalmanFilter.
Q = 0.2 * np.eye(4) # Process noise array
R = 0.5 * np.eye(4) # Measurement noise array
kf = ps5.KalmanFilter(template_loc['x'], template_loc['y'])
save_frames = {
10: os.path.join(output_dir, 'ps5-1-b-1.png'),
30: os.path.join(output_dir, 'ps5-1-b-2.png'),
59: os.path.join(output_dir, 'ps5-1-b-3.png'),
99: os.path.join(output_dir, 'ps5-1-b-4.png')
}
run_kalman_filter(kf, os.path.join(input_dir, "circle"), NOISE_2,
"matching", save_frames, template_loc)
def part_1c():
print("Part 1c")
init_pos = {'x': 311, 'y': 217}
# Define process and measurement arrays if you want to use other than the
# default. Pass them to KalmanFilter.
Q = 0.02 * np.eye(4) # Process noise array
R = 0.05 * np.eye(4) # Measurement noise array
kf = ps5.KalmanFilter(init_pos['x'], init_pos['y'])
save_frames = {
10: os.path.join(output_dir, 'ps5-1-c-1.png'),
33: os.path.join(output_dir, 'ps5-1-c-2.png'),
84: os.path.join(output_dir, 'ps5-1-c-3.png'),
159: os.path.join(output_dir, 'ps5-1-c-4.png')
}
run_kalman_filter(kf, os.path.join(input_dir, "walking"), NOISE_1, "hog",
save_frames)
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.
"""
template_loc1 = {'y': 180, 'x': 75, 'w': 50, 'h': 160}
template_loc2 = {'y': 200, 'x': 300, 'w': 35, 'h': 100}
# Define process and measurement arrays if you want to use other than the
# default. Pass them to KalmanFilter.
Q = 0.5 * np.eye(4) # Process noise array
R = 0.5 * np.eye(2) # Measurement noise array
kf1 = ps5.KalmanFilter(template_loc1['x'], template_loc1['y'])
kf2 = ps5.KalmanFilter(template_loc2['x'], template_loc2['y'])
#Params
noise = NOISE_2
sensor = "matching"
imgs_dir = os.path.join(input_dir, "TUD-Campus")
save_frames = {
29: os.path.join(output_dir, 'ps5-5-a-1.png'),
55: os.path.join(output_dir, 'ps5-5-a-2.png'),
70: os.path.join(output_dir, 'ps5-5-a-3.png')
}
# RUN KF
imgs_list = [
f for f in os.listdir(imgs_dir) if f[0] != '.' and f.endswith('.jpg')
]
imgs_list.sort()
frame_num = 0
if sensor == "matching":
frame = cv2.imread(os.path.join(imgs_dir, imgs_list[0]))
template1 = frame[template_loc1['y']:template_loc1['y'] +
template_loc1['h'],
template_loc1['x']:template_loc1['x'] +
template_loc1['w']]
template2 = frame[template_loc2['y']:template_loc2['y'] +
template_loc2['h'],
template_loc2['x']:template_loc2['x'] +
template_loc2['w']]
else:
raise ValueError("Unknown sensor name. Choose between 'hog' or "
"'matching'")
for img in imgs_list:
frame = cv2.imread(os.path.join(imgs_dir, img))
if frame_num == 29:
template_loc3 = {'y': 180, 'x': 0, 'w': 50, 'h': 150}
template3 = frame[template_loc3['y']:template_loc3['y'] +
template_loc3['h'],
template_loc3['x']:template_loc3['x'] +
template_loc3['w']]
kf3 = ps5.KalmanFilter(template_loc3['x'], template_loc3['y'], Q,
R)
# Sensor
if sensor == "matching":
if frame_num < 60:
corr_map1 = cv2.matchTemplate(frame, template1, cv2.TM_SQDIFF)
z_y1, z_x1 = np.unravel_index(np.argmin(corr_map1),
corr_map1.shape)
z_w1 = template_loc1['w']
z_h1 = template_loc1['h']
z_x1 += z_w1 // 2 + np.random.normal(0, noise['x'])
z_y1 += z_h1 // 2 + np.random.normal(0, noise['y'])
if frame_num < 45:
corr_map2 = cv2.matchTemplate(frame, template2, cv2.TM_SQDIFF)
z_y2, z_x2 = np.unravel_index(np.argmin(corr_map2),
corr_map2.shape)
z_w2 = template_loc2['w']
z_h2 = template_loc2['h']
z_x2 += z_w2 // 2 + np.random.normal(0, noise['x'])
z_y2 += z_h2 // 2 + np.random.normal(0, noise['y'])
if frame_num > 28:
corr_map3 = cv2.matchTemplate(frame, template3, cv2.TM_SQDIFF)
z_y3, z_x3 = np.unravel_index(np.argmin(corr_map3),
corr_map3.shape)
z_w3 = template_loc3['w']
z_h3 = template_loc3['h']
z_x3 += z_w3 // 2 + np.random.normal(0, noise['x'])
z_y3 += z_h3 // 2 + np.random.normal(0, noise['y'])
x1, y1 = kf1.process(z_x1, z_y1)
x2, y2 = kf2.process(z_x2, z_y2)
if frame_num > 28:
x3, y3 = kf3.process(z_x3, z_y3)
if False: # For debugging, it displays every frame
out_frame = frame.copy()
cv2.circle(out_frame, (int(z_x), int(z_y)), 20, (0, 0, 255), 2)
cv2.circle(out_frame, (int(x), int(y)), 10, (255, 0, 0), 2)
cv2.rectangle(out_frame,
(int(z_x) - z_w // 2, int(z_y) - z_h // 2),
(int(z_x) + z_w // 2, int(z_y) + z_h // 2),
(0, 0, 255), 2)
cv2.imshow('Tracking', out_frame)
cv2.waitKey(1)
# Render and save output, if indicated
if frame_num in save_frames:
print('Working on frame {}'.format(frame_num))
frame_out = frame.copy()
if frame_num < 60:
cv2.circle(frame_out, (int(x1), int(y1)), 10, (255, 0, 0), 2)
cv2.rectangle(frame_out,
(int(z_x1) - z_w1 // 2, int(z_y1) - z_h1 // 2),
(int(z_x1) + z_w1 // 2, int(z_y1) + z_h1 // 2),
(0, 0, 255), 2)
if frame_num < 45:
cv2.circle(frame_out, (int(x2), int(y2)), 10, (255, 0, 0), 2)
cv2.rectangle(frame_out,
(int(z_x2) - z_w2 // 2, int(z_y2) - z_h2 // 2),
(int(z_x2) + z_w2 // 2, int(z_y2) + z_h2 // 2),
(0, 0, 255), 2)
if frame_num >= 28:
cv2.circle(frame_out, (int(x3), int(y3)), 10, (255, 0, 0), 2)
cv2.rectangle(frame_out,
(int(z_x3) - z_w3 // 2, int(z_y3) - z_h3 // 2),
(int(z_x3) + z_w3 // 2, int(z_y3) + z_h3 // 2),
(0, 0, 255), 2)
cv2.imwrite(save_frames[frame_num], frame_out)
# Update frame number
frame_num += 1
