File size: 17,983 Bytes
6eb4316 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 | import numpy as np
import h5py
import logging
from dataclasses import dataclass
from typing import List, Tuple, Optional, Literal, Union
import os
import glob
import tyro
import pyzed.sl as sl
def setup_logging(name):
logging.basicConfig(level=logging.INFO, format='%(name)s - %(levelname)s - %(message)s')
return logging.getLogger(name)
log = setup_logging("Sync")
def _load_h5_dataset_1d(h5_path: str, dataset: str) -> np.ndarray:
"""
Load a 1D dataset from an HDF5 file.
"""
with h5py.File(h5_path, 'r') as f:
if dataset not in f:
raise ValueError(f"Dataset not found in HDF5 file: {dataset!r} (file: {h5_path})")
dset = f[dataset]
# Many recorders preallocate more slots than actual frames; honor num_frames attr if present.
num_frames_attr = f.attrs.get("num_frames")
if isinstance(num_frames_attr, (int, np.integer)) and 0 < num_frames_attr <= dset.shape[0]:
arr = dset[: int(num_frames_attr)]
else:
arr = dset[:]
arr = np.asarray(arr, dtype=np.float64)
if arr.ndim != 1:
raise ValueError(f"Expected 1D dataset {dataset!r}, got shape {arr.shape} (file: {h5_path})")
# Guard against preallocated-but-unwritten zeros (should never happen with time.time()*1000).
mask = arr > 0
return arr[mask]
def load_camera_timestamps_h5(camera_h5_path: str) -> np.ndarray:
"""
Load camera timestamps from HDF5 and remove the last one.
This ensures the number of timestamps is always safely within the SVO frame count,
avoiding off-by-one errors that can occur with SVO files.
Expected dataset: 'timestamps_ms' (milliseconds since Unix epoch).
Returns: timestamps in milliseconds, shape (N-1,)
"""
timestamps = _load_h5_dataset_1d(camera_h5_path, "timestamps_ms")
# Always remove the last timestamp to be safe
if len(timestamps) > 0:
timestamps = timestamps[:-1]
return timestamps
def load_radar_timestamps_h5(radar_h5_path: str) -> np.ndarray:
"""
Load radar timestamps from `src/radarRecorder.py` output HDF5.
Expected dataset: 'timestamps_ms' (milliseconds since Unix epoch).
"""
return _load_h5_dataset_1d(radar_h5_path, "timestamps_ms")
def load_dji_timestamps_h5(dji_h5_path: str) -> np.ndarray:
"""
Load DJI timestamps from `src/djiRecorder.py` output HDF5.
Expected dataset: 'timestamps_ms' (milliseconds since Unix epoch).
"""
return _load_h5_dataset_1d(dji_h5_path, "timestamps_ms")
@dataclass(frozen=True)
class SyncResult:
"""
Output of 2-way timestamp synchronization (camera + radar).
- `pairs`: list of (camera_frame_idx, radar_frame_idx)
- `diffs_ms`: signed difference in ms for each pair: camera_ts - radar_ts
"""
pairs: List[Tuple[int, int]]
diffs_ms: np.ndarray
camera_count: int
radar_count: int
tolerance_ms: float
@dataclass(frozen=True)
class SyncResult3Way:
"""
Output of 3-way timestamp synchronization (radar + ZED depth + DJI RGB).
- `triples`: list of (radar_frame_idx, zed_frame_idx, dji_frame_idx)
- `diffs_radar_zed_ms`: signed difference radar_ts - zed_ts
- `diffs_radar_dji_ms`: signed difference radar_ts - dji_ts
"""
triples: List[Tuple[int, int, int]]
diffs_radar_zed_ms: np.ndarray
diffs_radar_dji_ms: np.ndarray
radar_count: int
zed_count: int
dji_count: int
tolerance_ms: float
def save_sync_csv(out_csv_path: str, result: Union[SyncResult, SyncResult3Way]) -> None:
"""
Save sync pairs/triples to CSV.
For SyncResult (2-way):
camera_frame_idx,radar_frame_idx,diff_ms
For SyncResult3Way (3-way):
radar_frame_idx,zed_frame_idx,dji_frame_idx,diff_radar_zed_ms,diff_radar_dji_ms
"""
os.makedirs(os.path.dirname(out_csv_path) or ".", exist_ok=True)
if isinstance(result, SyncResult3Way):
# 3-way sync
radar_idx = np.array([t[0] for t in result.triples], dtype=np.int64)
zed_idx = np.array([t[1] for t in result.triples], dtype=np.int64)
dji_idx = np.array([t[2] for t in result.triples], dtype=np.int64)
arr = np.column_stack([
radar_idx,
zed_idx,
dji_idx,
result.diffs_radar_zed_ms.astype(np.float64, copy=False),
result.diffs_radar_dji_ms.astype(np.float64, copy=False)
])
header = "radar_frame_idx,zed_frame_idx,dji_frame_idx,diff_radar_zed_ms,diff_radar_dji_ms"
fmt = ["%d", "%d", "%d", "%.6f", "%.6f"]
else:
# 2-way sync (backward compatibility)
cam_idx = np.array([p[0] for p in result.pairs], dtype=np.int64)
rad_idx = np.array([p[1] for p in result.pairs], dtype=np.int64)
arr = np.column_stack([cam_idx, rad_idx, result.diffs_ms.astype(np.float64, copy=False)])
header = "camera_frame_idx,radar_frame_idx,diff_ms"
fmt = ["%d", "%d", "%.6f"]
np.savetxt(out_csv_path, arr, fmt=fmt, delimiter=",", header=header, comments="")
def _pick_latest_file(directory: str, pattern: str) -> str:
"""
Pick the latest file (by lexicographic sort) matching a pattern in a directory.
This works for our timestamped filenames like:
- camera_timestamps_YYYYmmdd_HHMMSS.h5
- radar_YYYYmmdd_HHMMSS.h5
"""
matches = glob.glob(os.path.join(directory, pattern))
if not matches:
raise FileNotFoundError(f"No files matching {pattern!r} in directory: {directory}")
matches.sort()
return matches[-1]
def _pick_latest_file_multi_ext(directory: str, patterns: list[str]) -> str:
"""
Pick the latest file matching any of multiple patterns.
Useful for finding files with different extensions (e.g., .mkv or .mp4).
Example:
path = _pick_latest_file_multi_ext("data", ["dji_*.mkv", "dji_*.mp4"])
"""
all_matches = []
for pattern in patterns:
matches = glob.glob(os.path.join(directory, pattern))
all_matches.extend(matches)
if not all_matches:
patterns_str = " or ".join(patterns)
raise FileNotFoundError(f"No files matching {patterns_str} in directory: {directory}")
all_matches.sort()
return all_matches[-1]
def cli(
directory: Optional[str] = None,
camera_timestamps_h5: Optional[str] = None,
radar_h5: Optional[str] = None,
svo_path: Optional[str] = None,
out_csv: str = os.path.join("data", "sync_pairs.csv"),
tolerance_ms: float = 50.0,
enforce_one_to_one: bool = True,
) -> None:
"""
CLI wrapper for timestamp synchronization.
Usage patterns:
1) Directory mode (auto-pick latest files):
- Provide `directory` and omit `camera_timestamps_h5` / `radar_h5` / `svo_path`.
2) File mode (explicit paths):
- Provide BOTH `camera_timestamps_h5` and `radar_h5`, optionally `svo_path`.
"""
if directory is not None:
if camera_timestamps_h5 is not None or radar_h5 is not None or svo_path is not None:
raise ValueError("If 'directory' is provided, do not also pass explicit file paths.")
camera_timestamps_h5 = _pick_latest_file(directory, "camera_timestamps_*.h5")
radar_h5 = _pick_latest_file(directory, "radar_*.h5")
svo_path = _pick_latest_file(directory, "zed_*.svo2")
log.info(f"Auto-picked camera timestamps: {camera_timestamps_h5}")
log.info(f"Auto-picked radar file: {radar_h5}")
log.info(f"Auto-picked SVO: {svo_path}")
if camera_timestamps_h5 is None or radar_h5 is None:
raise ValueError(
"Provide either:\n"
"- directory=<folder containing camera_timestamps_*.h5 and radar_*.h5>, OR\n"
"- camera_timestamps_h5=<path> AND radar_h5=<path>."
)
result = synchronize_timestamps(
camera_timestamps_h5=camera_timestamps_h5,
radar_h5=radar_h5,
tolerance_ms=tolerance_ms,
enforce_one_to_one=enforce_one_to_one,
svo_path=svo_path,
)
save_sync_csv(out_csv, result)
log.info(f"Wrote sync CSV: {out_csv}")
def _nearest_unused_camera_index(
cam_ts: np.ndarray,
target_t: float,
used_camera: set[int],
start_idx: int,
) -> Optional[int]:
"""
Find nearest unused camera index to `target_t`.
`start_idx` is the insertion index from `np.searchsorted(cam_ts, target_t)`.
We expand outward until we find an unused camera frame.
"""
left = start_idx - 1
right = start_idx
while left >= 0 or right < len(cam_ts):
cand_left = left if left >= 0 else None
cand_right = right if right < len(cam_ts) else None
if cand_left is None and cand_right is None:
return None
best: Optional[int] = None
best_abs = float("inf")
for cand in (cand_left, cand_right):
if cand is None:
continue
if cand in used_camera:
continue
abs_diff = abs(float(cam_ts[cand] - target_t))
if abs_diff < best_abs:
best_abs = abs_diff
best = int(cand)
if best is not None:
return best
left -= 1
right += 1
return None
def synchronize_timestamps(
camera_timestamps_h5: str,
radar_h5: str,
tolerance_ms: float = 50.0,
enforce_one_to_one: bool = True,
strategy: Literal["nearest"] = "nearest",
svo_path: Optional[str] = None,
) -> SyncResult:
"""
Synchronize camera and radar timestamps.
Assumptions (true for the new recorders):
- Both timestamp streams are in **milliseconds since Unix epoch** (from `time.time()*1000`).
- Each stream is **monotonically increasing**.
Parameters:
- svo_path: Optional path to SVO2 file. If provided, validates that camera timestamps
don't exceed the actual number of frames in the SVO.
Returns pairs and per-pair signed diffs: Δt = t_cam - t_radar.
"""
if strategy != "nearest":
raise ValueError(f"Unsupported strategy: {strategy!r}")
cam_ts = load_camera_timestamps_h5(camera_timestamps_h5)
rad_ts = load_radar_timestamps_h5(radar_h5)
log.info(f"Camera timestamps: {len(cam_ts)} (last timestamp already removed for SVO safety)")
# Ensure float for arithmetic (HDF5 may store float64 already, but keep consistent)
cam_ts = cam_ts.astype(np.float64, copy=False)
rad_ts = rad_ts.astype(np.float64, copy=False)
# Reverse logic: iterate radar frames and pick nearest camera frame.
# Goal: keep (as many as possible) radar frames, since radar is the lower FPS stream.
pairs: List[Tuple[int, int]] = []
diffs: List[float] = []
used_camera: set[int] = set()
for rad_idx, r_t in enumerate(rad_ts):
insert_idx = int(np.searchsorted(cam_ts, r_t))
if enforce_one_to_one:
cam_idx = _nearest_unused_camera_index(cam_ts, float(r_t), used_camera, insert_idx)
else:
# Nearest of the two immediate neighbors (can reuse camera frames).
cand0 = insert_idx - 1 if insert_idx > 0 else None
cand1 = insert_idx if insert_idx < len(cam_ts) else None
best: Optional[int] = None
best_abs = float("inf")
for cand in (cand0, cand1):
if cand is None:
continue
abs_diff = abs(float(cam_ts[cand] - r_t))
if abs_diff < best_abs:
best_abs = abs_diff
best = int(cand)
cam_idx = best
if cam_idx is None:
# No camera frames at all (or no unused ones left).
break
signed = float(cam_ts[cam_idx] - r_t)
abs_diff = abs(signed)
if abs_diff > tolerance_ms:
continue
pairs.append((int(cam_idx), int(rad_idx)))
diffs.append(signed)
if enforce_one_to_one:
used_camera.add(int(cam_idx))
diffs_ms = np.array(diffs, dtype=np.float64)
log.info(
f"Synchronized {len(pairs)} pairs | "
f"camera={len(cam_ts)} radar={len(rad_ts)} | tol={tolerance_ms}ms | "
f"one_to_one={enforce_one_to_one}"
)
return SyncResult(
pairs=pairs,
diffs_ms=diffs_ms,
camera_count=int(len(cam_ts)),
radar_count=int(len(rad_ts)),
tolerance_ms=float(tolerance_ms),
)
if __name__ == "__main__":
tyro.cli(cli)
def synchronize_timestamps_3way(
radar_h5: str,
zed_timestamps_h5: str,
dji_timestamps_h5: str,
tolerance_ms: float = 50.0,
enforce_one_to_one: bool = True,
svo_path: Optional[str] = None,
dji_video_path: Optional[str] = None,
) -> SyncResult3Way:
"""
Synchronize 3 timestamp streams: radar (reference), ZED depth, DJI RGB.
Strategy:
- Radar is the lowest FPS stream (typically ~10 Hz), so we use it as the reference.
- For each radar frame, find the nearest ZED frame and nearest DJI frame.
- Both must be within tolerance_ms of the radar timestamp.
Parameters:
- radar_h5: Path to radar HDF5 file with timestamps_ms
- zed_timestamps_h5: Path to ZED camera timestamps HDF5
- dji_timestamps_h5: Path to DJI timestamps HDF5
- tolerance_ms: Maximum time difference in milliseconds
- enforce_one_to_one: If True, each ZED/DJI frame can only be used once
- svo_path: Optional path to SVO2 file for frame count validation
- dji_video_path: Optional path to DJI video for frame count validation
Returns:
- SyncResult3Way with triples (radar_idx, zed_idx, dji_idx)
"""
# Load all timestamps
radar_ts = load_radar_timestamps_h5(radar_h5)
zed_ts = load_camera_timestamps_h5(zed_timestamps_h5)
dji_ts = load_dji_timestamps_h5(dji_timestamps_h5)
log.info(f"ZED timestamps: {len(zed_ts)} (last timestamp already removed for SVO safety)")
# Ensure float64 for arithmetic
radar_ts = radar_ts.astype(np.float64, copy=False)
zed_ts = zed_ts.astype(np.float64, copy=False)
dji_ts = dji_ts.astype(np.float64, copy=False)
log.info(f"Timestamp ranges:")
log.info(f" Radar: {len(radar_ts)} frames, {radar_ts[0]:.2f} - {radar_ts[-1]:.2f} ms")
log.info(f" ZED: {len(zed_ts)} frames, {zed_ts[0]:.2f} - {zed_ts[-1]:.2f} ms")
log.info(f" DJI: {len(dji_ts)} frames, {dji_ts[0]:.2f} - {dji_ts[-1]:.2f} ms")
# Iterate through radar frames (lowest FPS, reference stream)
triples = []
diffs_radar_zed = []
diffs_radar_dji = []
used_zed = set()
used_dji = set()
for radar_idx, r_t in enumerate(radar_ts):
# Find nearest ZED frame
zed_insert_idx = int(np.searchsorted(zed_ts, r_t))
if enforce_one_to_one:
zed_idx = _nearest_unused_camera_index(zed_ts, float(r_t), used_zed, zed_insert_idx)
else:
cand0 = zed_insert_idx - 1 if zed_insert_idx > 0 else None
cand1 = zed_insert_idx if zed_insert_idx < len(zed_ts) else None
zed_idx = None
best_abs = float("inf")
for cand in (cand0, cand1):
if cand is None:
continue
abs_diff = abs(float(zed_ts[cand] - r_t))
if abs_diff < best_abs:
best_abs = abs_diff
zed_idx = int(cand)
if zed_idx is None:
continue
zed_diff = float(r_t - zed_ts[zed_idx])
if abs(zed_diff) > tolerance_ms:
continue
# Find nearest DJI frame
dji_insert_idx = int(np.searchsorted(dji_ts, r_t))
if enforce_one_to_one:
dji_idx = _nearest_unused_camera_index(dji_ts, float(r_t), used_dji, dji_insert_idx)
else:
cand0 = dji_insert_idx - 1 if dji_insert_idx > 0 else None
cand1 = dji_insert_idx if dji_insert_idx < len(dji_ts) else None
dji_idx = None
best_abs = float("inf")
for cand in (cand0, cand1):
if cand is None:
continue
abs_diff = abs(float(dji_ts[cand] - r_t))
if abs_diff < best_abs:
best_abs = abs_diff
dji_idx = int(cand)
if dji_idx is None:
continue
dji_diff = float(r_t - dji_ts[dji_idx])
if abs(dji_diff) > tolerance_ms:
continue
# Both ZED and DJI within tolerance - add triple
triples.append((int(radar_idx), int(zed_idx), int(dji_idx)))
diffs_radar_zed.append(zed_diff)
diffs_radar_dji.append(dji_diff)
if enforce_one_to_one:
used_zed.add(int(zed_idx))
used_dji.add(int(dji_idx))
diffs_radar_zed_ms = np.array(diffs_radar_zed, dtype=np.float64)
diffs_radar_dji_ms = np.array(diffs_radar_dji, dtype=np.float64)
log.info(
f"Synchronized {len(triples)} triples | "
f"radar={len(radar_ts)} zed={len(zed_ts)} dji={len(dji_ts)} | "
f"tol={tolerance_ms}ms | one_to_one={enforce_one_to_one}"
)
return SyncResult3Way(
triples=triples,
diffs_radar_zed_ms=diffs_radar_zed_ms,
diffs_radar_dji_ms=diffs_radar_dji_ms,
radar_count=int(len(radar_ts)),
zed_count=int(len(zed_ts)),
dji_count=int(len(dji_ts)),
tolerance_ms=float(tolerance_ms),
)
|