/
FusionVisionV0.3.py
248 lines (196 loc) · 12.9 KB
/
FusionVisionV0.3.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
import cv2
import numpy as np
import argparse
from FastSAM.fastsam import FastSAM, FastSAMPrompt
from ultralytics import YOLO
import open3d as o3d
import pyrealsense2 as rs
from utils import perform_yolo_inference
import csv
def parse_args():
parser = argparse.ArgumentParser(description="Fusion Vision Object Detection and 3D Reconstruction")
parser.add_argument("--yolo_weight", type=str, required=True,
help="Path to the YOLO weights file (e.g yolo_train/runs/detect/train/weights/best.pt)")
parser.add_argument("--fastsam_weight", type=str, choices=['FastSAM-x.pt', 'FastSAM-s.pt'], default='FastSAM-x.pt',
help="Choose the FastSAM autodownloadable weight files ('FastSAM-x.pt' or 'FastSAM-s.pt')")
parser.add_argument("--show_yolo", action='store_true',
help="Show cv2 window with YOLO detection (default True)")
parser.add_argument("--show_fastsam", action='store_true',
help="Show cv2 window with FastSAM detection (default True)")
parser.add_argument("--show_mask", action='store_true',
help="Show a window with the estimated binary masks (default True)")
parser.add_argument("--confidence_threshold", type=float, default=0.7,
help="Set the confidence threshold for YOLO detection (default: 0.7)")
parser.add_argument("--conf", type=float, default=0.4,
help="Set the confidence threshold for the FastSAM model (default: 0.4)")
parser.add_argument("--iou", type=float, default=0.9,
help="Set the IoU threshold for non-maximum suppression (default: 0.9)")
parser.add_argument("--show_3dbbox", action='store_true',
help="Show in open3D window the 3D bounding box (default: True)")
return parser.parse_args()
def FusionVision(args):
pipeline = rs.pipeline()
config = rs.config()
config.enable_stream(rs.stream.color, 640, 480, rs.format.bgr8, 30)
config.enable_stream(rs.stream.depth, 640, 480, rs.format.z16, 30)
pipeline.start(config)
yolo_model = YOLO(args.yolo_weight)
fastsam_model_path = args.fastsam_weight
fastsam_model = FastSAM(fastsam_model_path)
visualizer = o3d.visualization.Visualizer()
visualizer.create_window("Point Cloud Viewer", width=640, height=480, visible=True)
pcd = o3d.geometry.PointCloud()
# Transformation matrix for flipping the point cloud upside down and left to right
flip_matrix = np.array([[1, 0, 0, 0],
[0, -1, 0, 0],
[0, 0, -1, 0],
[0, 0, 0, 1]])
if args.show_fastsam:
cv2.namedWindow('FastSAM Inference', cv2.WINDOW_NORMAL)
if args.show_mask:
cv2.namedWindow('Annotation Mask', cv2.WINDOW_NORMAL)
if args.show_mask:
cv2.namedWindow('YOLO Inference', cv2.WINDOW_NORMAL)
# Create an align object
align_to = rs.align(rs.stream.color)
coordinates_list=[]
try:
while True:
# Capture frames from RealSense camera
frames = pipeline.wait_for_frames()
color_frame = frames.get_color_frame()
aligned_frames = align_to.process(frames)
aligned_depth_frame = aligned_frames.get_depth_frame()
color_image = np.asanyarray(color_frame.get_data())
depth_image = np.asanyarray(aligned_depth_frame.get_data())
# Perform YOLO inference on the color image
detections, predicted_boxes = perform_yolo_inference(color_image, yolo_model, confidence_threshold=args.confidence_threshold)
# Create a list to store bounding box lines
bounding_box_lines = []
if len(predicted_boxes) > 0:
bounding_boxes = [list(map(int, box[:4])) for box in predicted_boxes]
fastsam_results = fastsam_model(color_image, device='cuda:0', retina_masks=True, imgsz=640, conf=args.conf,
iou=args.iou)
if fastsam_results:
prompt_process = FastSAMPrompt(color_image, fastsam_results, device='cuda:0')
ann = prompt_process.box_prompt(bboxes=bounding_boxes)
img_with_annotations = prompt_process.plot_to_result(annotations=ann)
# Display FastSAM results in another window
if args.show_fastsam:
cv2.imshow('FastSAM Inference', img_with_annotations)
ann_mask = np.array(ann).astype(np.uint8)
ann_mask_overlay = np.sum(ann_mask, axis=0)
# Normalize the overlay to be in the range [0, 1] and convert to uint8
ann_mask_overlay_normalized = (ann_mask_overlay / np.max(ann_mask_overlay) * 255).astype(np.uint8)
if args.show_mask:
cv2.imshow('Annotation Mask', ann_mask_overlay_normalized)
ann_mask_overlay_uint8 = ann_mask_overlay.astype(np.uint8)
# Erode the annotation mask (to avoid reconstructing in 3D some background)
eroded_ann_mask = cv2.erode(ann_mask_overlay_uint8, kernel=np.ones((20, 20), np.uint8), iterations=1)
isolated_depth = np.where((eroded_ann_mask > 0) & (depth_image < 1000), depth_image, np.nan)
non_nan_points = np.argwhere(~np.isnan(isolated_depth))
non_nan_depth_values = isolated_depth[non_nan_points[:, 0], non_nan_points[:, 1]]
depth_scale = 1
pcd.points = o3d.utility.Vector3dVector(
np.column_stack([non_nan_points[:, 1], non_nan_points[:, 0], non_nan_depth_values * depth_scale])
)
pcd_outlier = pcd.voxel_down_sample(voxel_size=2)
denoised_pcd, _ = pcd_outlier.remove_statistical_outlier(nb_neighbors=300,
std_ratio=2.0)
for box in bounding_boxes:
x_min, y_min, x_max, y_max = box
# Extract the region of interest from the denoised point cloud
roi_points = np.asarray(denoised_pcd.points)
roi_points = roi_points[(roi_points[:, 0] >= x_min) & (roi_points[:, 0] <= x_max) &
(roi_points[:, 1] >= y_min) & (roi_points[:, 1] <= y_max)]
# Compute the bounding box dimensions based on the region of interest
bbox_lines = o3d.geometry.LineSet()
bbox_lines.points = o3d.utility.Vector3dVector([
[np.min(roi_points[:, 0]), np.min(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.min(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.min(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.max(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.max(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.max(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.max(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.min(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.min(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.min(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.min(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.max(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.max(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.max(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.max(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.min(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.min(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.min(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.min(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.min(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.max(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.max(roi_points[:, 0]), np.max(roi_points[:, 1]), np.max(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.max(roi_points[:, 1]), np.min(roi_points[:, 2])],
[np.min(roi_points[:, 0]), np.max(roi_points[:, 1]), np.max(roi_points[:, 2])],
])
bbox_lines.lines = o3d.utility.Vector2iVector([
[0, 1], [1, 2], [2, 3], [3, 0],
[4, 5], [5, 6], [6, 7], [7, 4],
[0, 7], [1, 6], [2, 5], [3, 4],
[8, 9], [9, 10], [10, 11], [11, 8],
[12, 13], [13, 14], [14, 15], [15, 12],
[8, 15], [9, 14], [10, 13], [11, 12],
[16, 17], [17, 18], [18, 19], [19, 16],
[20, 21], [21, 22], [22, 23], [23, 20],
[16, 23], [17, 22], [18, 21], [19, 20]
])
bbox_lines.colors = o3d.utility.Vector3dVector([[1, 0, 0] for _ in range(len(bbox_lines.lines))])
bounding_box_lines.append(bbox_lines)
denoised_pcd.transform(flip_matrix)
visualizer.clear_geometries()
visualizer.add_geometry(denoised_pcd)
if args.show_3dbbox:
for bbox_lines in bounding_box_lines:
bbox_lines.transform(flip_matrix)
visualizer.add_geometry(bbox_lines)
center = np.mean(np.asarray(bbox_lines.points), axis=0)
coordinates_list.append(center)
# length of the coordinate axes
axis_length = 50
# Create coordinate frame mesh
coordinate_system = o3d.geometry.TriangleMesh.create_coordinate_frame(size=axis_length, origin=center)
visualizer.add_geometry(coordinate_system)
visualizer.poll_events()
visualizer.update_renderer()
csv_file_path = 'object_coordinates.csv'
# Write the coordinates to the CSV file
with open(csv_file_path, mode='w', newline='') as file:
writer = csv.writer(file)
# Write header if needed
writer.writerow(['X', 'Y', 'Z'])
# Write each set of coordinates
for coordinates in coordinates_list:
writer.writerow(coordinates)
# if YOLO window = true
if args.show_yolo:
for detection in detections:
x1, y1, x2, y2 = detection['bounding_box']
confidence = detection['confidence']
class_name = detection['class_name']
cv2.rectangle(color_image, (x1, y1), (x2, y2), (255, 255, 0))
org = (x1, y1 - 10)
font = cv2.FONT_HERSHEY_SIMPLEX
font_scale = 1
color = (255, 255, 255)
thickness = 2
cv2.putText(color_image, f"{class_name}: {confidence}", org, font, font_scale, color, thickness)
cv2.imshow('YOLO Inference', color_image)
# Check for key press to exit the loop (press 'q' to quit)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
finally:
# Stop streaming
pipeline.stop()
visualizer.destroy_window()
cv2.destroyAllWindows()
if __name__ == "__main__":
args = parse_args()
FusionVision(args)