Skip to content

Multi-Camera Video Synchronization with ISOBMFF

Extract Rhombus millisecond timestamps embedded in ISOBMFF video containers to align frames precisely across multiple cameras for forensic analysis.

On this page

Rhombus writes a millisecond-precision UTC timestamp into every recorded .mp4/.m4v segment. The standard timing fields in ISOBMFF (ISO Base Media File Format) containers have one-second resolution, which isn't enough to line up the same moment across several cameras, audio and sensor data. This guide explains where the timestamp is stored and how to read it.

Understanding ISOBMFF and the "free" atom

The ISOBMFF standard (ISO/IEC 14496-12) is the container format underlying .mp4, .m4v, .mov, and segment formats such as fragmented MP4. It is built from boxes (atoms): self-contained data units identified by a four-character code, such as moov, mdat and free.

Standard timestamps in ISOBMFF, such as creation_time in the mvhd box, have one-second resolution: enough for most media workflows, but not for multi-camera synchronization or correlating fast events. Rhombus adds a millisecond-precision timestamp inside a free atom, which players ignore, so the files remain standard-compliant.

The Rhombus custom timestamp

When Rhombus segments video (and if pulling actual segments, not a live transport stream), a custom metadata signature is written into the free box:

  1. Box Type: free

    In ISOBMFF, free is normally a placeholder box containing unused space. Rhombus repurposes this to carry timestamp metadata.

  2. Signature: rhom

    The first 4 bytes are the ASCII string rhom—identifying it as Rhombus-specific data.

  3. Millisecond Timestamp

    The next 8 bytes are a 64-bit integer representing the start time of the video content, measured in milliseconds since the Unix epoch (UTC).

Binary layout example:

[ free box length ][ 'free' ][ 'rhom' ][ 8-byte timestamp (ms since epoch) ]

Why this is important for developers

This design choice unlocks precise synchronization capabilities across multiple use cases:

Multi-Camera Alignment

Align video feeds from different cameras to within 1 ms for coordinated monitoring

Sensor Fusion

Merge video with IoT sensor data (access control events, environmental readings)

Forensic Accuracy

Reconstruct events down to sub-second intervals in investigations

Reduced Drift

Avoid errors that accumulate when relying solely on client system clocks or NTP sync

Retrieving the timestamp

Parsing ISOBMFF files

You can use open-source libraries to read the free box from an .mp4/.m4v segment and check for the rhom signature.

Implementation examples

Python Implementationpython
import struct
import datetime

def extract_rhombus_timestamp(file_path):
    """
    Extract millisecond-precision timestamp from Rhombus video segment.

    Args:
        file_path: Path to the .mp4 or .m4v file

    Returns:
        tuple: (timestamp_ms, datetime object) or (None, None) if not found
    """
    with open(file_path, "rb") as f:
        data = f.read()

    # Find the 'free' box
    idx = data.find(b'free')
    if idx == -1:
        return None, None

    # Search for 'rhom' signature after 'free'
    rhom_idx = data.find(b'rhom', idx)
    if rhom_idx == -1:
        return None, None

    # Read the next 8 bytes after 'rhom' (big-endian 64-bit integer)
    timestamp_bytes = data[rhom_idx + 4 : rhom_idx + 12]
    timestamp_ms = int.from_bytes(timestamp_bytes, byteorder="big")

    # Convert to human-readable UTC time
    timestamp_dt = datetime.datetime.fromtimestamp(timestamp_ms / 1000.0, tz=datetime.timezone.utc)

    return timestamp_ms, timestamp_dt

# Example usage:
timestamp_ms, timestamp_dt = extract_rhombus_timestamp("video_segment.mp4")

if timestamp_ms:
    print(f"Timestamp (ms since epoch): {timestamp_ms}")
    print(f"UTC Time: {timestamp_dt}")
    print(f"ISO Format: {timestamp_dt.isoformat()}")
else:
    print("Rhombus timestamp not found in file")

# Sample output:
# Timestamp (ms since epoch): 1722945678123
# UTC Time: 2024-08-06 15:21:18.123000
# ISO Format: 2024-08-06T15:21:18.123000

Advanced Python Usage

Multi-Segment Processingpython
import os
import glob
from datetime import datetime

class RhombusTimestampExtractor:
    """Extract and manage timestamps from multiple video segments."""

    def __init__(self):
        self.timestamps = []

    def process_directory(self, directory_path, pattern="*.mp4"):
        """Process all video files in a directory."""
        files = glob.glob(os.path.join(directory_path, pattern))

        for file_path in sorted(files):
            timestamp_ms, timestamp_dt = extract_rhombus_timestamp(file_path)

            if timestamp_ms:
                self.timestamps.append({
                    'file': os.path.basename(file_path),
                    'timestamp_ms': timestamp_ms,
                    'datetime': timestamp_dt
                })

    def get_timeline(self):
        """Get chronologically sorted list of segments."""
        return sorted(self.timestamps, key=lambda x: x['timestamp_ms'])

    def find_segment_at_time(self, target_datetime):
        """Find the video segment containing a specific time."""
        target_ms = int(target_datetime.timestamp() * 1000)

        for segment in self.get_timeline():
            if segment['timestamp_ms'] <= target_ms:
                closest = segment
            else:
                break

        return closest

# Usage example
extractor = RhombusTimestampExtractor()
extractor.process_directory("/path/to/video/segments")

# Find segment at specific time
target = datetime(2024, 8, 6, 15, 21, 0)
segment = extractor.find_segment_at_time(target)
print(f"Segment at {target}: {segment['file']}")

Real-world use cases

Multi-camera event reconstruction

Synchronize footage from multiple cameras to reconstruct security incidents:

Event Timeline Reconstructionpython
from datetime import datetime, timedelta

class EventReconstructor:
    def __init__(self):
        self.camera_segments = {}  # cameraId -> list of segments

    def add_camera_footage(self, camera_id, segment_files):
        """Add video segments from a specific camera."""
        segments = []

        for file_path in segment_files:
            timestamp_ms, timestamp_dt = extract_rhombus_timestamp(file_path)
            if timestamp_ms:
                segments.append({
                    'file': file_path,
                    'start_time': timestamp_dt,
                    'timestamp_ms': timestamp_ms
                })

        self.camera_segments[camera_id] = sorted(
            segments,
            key=lambda x: x['timestamp_ms']
        )

    def reconstruct_event(self, event_time, window_seconds=30):
        """
        Find all camera segments within time window of an event.

        Args:
            event_time: datetime of the event
            window_seconds: seconds before/after event to include
        """
        window = timedelta(seconds=window_seconds)
        start_time = event_time - window
        end_time = event_time + window

        relevant_footage = {}

        for camera_id, segments in self.camera_segments.items():
            camera_clips = []

            for segment in segments:
                # Check if segment overlaps with event window
                if start_time <= segment['start_time'] <= end_time:
                    offset = (segment['start_time'] - event_time).total_seconds()
                    camera_clips.append({
                        'file': segment['file'],
                        'start_time': segment['start_time'],
                        'offset_from_event': offset
                    })

            if camera_clips:
                relevant_footage[camera_id] = camera_clips

        return relevant_footage

# Usage
reconstructor = EventReconstructor()
reconstructor.add_camera_footage('entrance', entrance_files)
reconstructor.add_camera_footage('lobby', lobby_files)
reconstructor.add_camera_footage('parking', parking_files)

# Reconstruct event at specific time
event_time = datetime(2024, 8, 6, 15, 21, 18)
footage = reconstructor.reconstruct_event(event_time, window_seconds=30)

print(f"Footage for event at {event_time}:")
for camera_id, clips in footage.items():
    print(f"\n{camera_id}:")
    for clip in clips:
        print(f"  - {clip['file']}")
        print(f"    Offset: {clip['offset_from_event']:.2f}s")

Sensor data correlation

Align video with access control or environmental sensor events:

Sensor-Video Correlationjavascript
class SensorVideoCorrelator {
    constructor() {
        this.videoTimestamps = [];
        this.sensorEvents = [];
    }

    /**
     * Add video segment with its timestamp
     */
    addVideoSegment(cameraId, timestamp, videoUrl) {
        this.videoTimestamps.push({
            cameraId,
            timestamp: timestamp.timestampMs,
            datetime: timestamp.utcTime,
            url: videoUrl
        });
    }

    /**
     * Add sensor event with timestamp
     */
    addSensorEvent(sensorId, eventType, timestampMs, data) {
        this.sensorEvents.push({
            sensorId,
            eventType,
            timestamp: timestampMs,
            datetime: new Date(timestampMs),
            data
        });
    }

    /**
     * Find video coverage for a sensor event
     */
    findVideoForEvent(eventTimestampMs, cameras = null) {
        const relevantVideos = this.videoTimestamps.filter(video => {
            // Video segment starts before or at event time
            const inTimeRange = video.timestamp <= eventTimestampMs;

            // Filter by camera if specified
            const inCameraList = !cameras || cameras.includes(video.cameraId);

            return inTimeRange && inCameraList;
        });

        // Get the most recent video before the event for each camera
        const latestByCamera = {};

        relevantVideos.forEach(video => {
            if (!latestByCamera[video.cameraId] ||
                video.timestamp > latestByCamera[video.cameraId].timestamp) {
                latestByCamera[video.cameraId] = video;
            }
        });

        return Object.values(latestByCamera);
    }

    /**
     * Generate correlation report
     */
    generateCorrelationReport() {
        return this.sensorEvents.map(event => {
            const videos = this.findVideoForEvent(event.timestamp);

            return {
                event: {
                    type: event.eventType,
                    sensor: event.sensorId,
                    time: event.datetime.toISOString(),
                    data: event.data
                },
                associatedVideos: videos.map(v => ({
                    camera: v.cameraId,
                    url: v.url,
                    offset: event.timestamp - v.timestamp
                }))
            };
        });
    }
}

// Usage example
const correlator = new SensorVideoCorrelator();

// Add door access event
correlator.addSensorEvent(
    'door-entrance-1',
    'ACCESS_GRANTED',
    1722945678123,
    { userId: 'user123', cardId: '12345' }
);

// Add corresponding video
correlator.addVideoSegment(
    'camera-entrance',
    { timestampMs: 1722945670000, utcTime: new Date(1722945670000) },
    'https://media.rhombussystems.com/segment1.mp4'
);

// Generate report
const report = correlator.generateCorrelationReport();
console.log(JSON.stringify(report, null, 2));

Best practices for integration

Validate the Signature

Always confirm the rhom tag before interpreting the following bytes as a timestamp. This prevents misinterpretation of unrelated data.

python
def is_valid_rhombus_timestamp(data, rhom_index):
    # Verify 'rhom' signature
    if data[rhom_index:rhom_index + 4] != b'rhom':
        return False

    # Verify sufficient data for timestamp
    if len(data) < rhom_index + 12:
        return False

    # Extract and validate timestamp range
    timestamp_bytes = data[rhom_index + 4:rhom_index + 12]
    timestamp_ms = int.from_bytes(timestamp_bytes, byteorder="big")

    # Sanity check: timestamp should be reasonable
    # (between 2015 and 2050)
    min_timestamp = 1420070400000  # Jan 1, 2015
    max_timestamp = 2524608000000  # Jan 1, 2050

    return min_timestamp <= timestamp_ms <= max_timestamp
Handle Time Zones Correctly

The timestamp is UTC-based. Convert it appropriately if your application needs local time.

python
import datetime as dt
import pytz

def convert_to_local_time(timestamp_ms, timezone='America/New_York'):
    # Create UTC datetime
    utc_dt = dt.datetime.fromtimestamp(timestamp_ms / 1000.0, tz=dt.timezone.utc)
    utc_dt = pytz.utc.localize(utc_dt)

    # Convert to local timezone
    local_tz = pytz.timezone(timezone)
    local_dt = utc_dt.astimezone(local_tz)

    return local_dt

# Usage
timestamp_ms = 1722945678123
local_time = convert_to_local_time(timestamp_ms, 'America/Los_Angeles')
print(f"Local time: {local_time}")
Use as Reference Clock

Combine the segment start timestamp with frame timestamps for frame-accurate alignment.

javascript
class FrameTimestampCalculator {
    constructor(segmentStartMs, frameRate) {
        this.segmentStartMs = segmentStartMs;
        this.frameRate = frameRate;
        this.frameDurationMs = 1000 / frameRate;
    }

    /**
     * Calculate absolute timestamp for a specific frame
     */
    getFrameTimestamp(frameNumber) {
        const offsetMs = frameNumber * this.frameDurationMs;
        return this.segmentStartMs + offsetMs;
    }

    /**
     * Find frame number for a specific timestamp
     */
    getFrameAtTimestamp(targetTimestampMs) {
        const offsetMs = targetTimestampMs - this.segmentStartMs;
        return Math.floor(offsetMs / this.frameDurationMs);
    }
}

// Usage
const segmentTimestamp = 1722945678123;
const calculator = new FrameTimestampCalculator(segmentTimestamp, 30);

// Get timestamp of frame 150
const frameTime = calculator.getFrameTimestamp(150);
console.log('Frame 150 timestamp:', new Date(frameTime));

// Find frame at specific time
const targetTime = 1722945683123;
const frameNum = calculator.getFrameAtTimestamp(targetTime);
console.log('Frame at target time:', frameNum);
Version for Future-Proofing

Store your parsing logic in a modular way in case Rhombus adds new metadata formats.

python
class RhombusMetadataParser:
    """Extensible parser for Rhombus metadata formats."""

    VERSION = "1.0"

    def __init__(self):
        self.parsers = {
            b'rhom': self._parse_v1_timestamp
        }

    def parse(self, file_path):
        """Parse metadata from video file."""
        with open(file_path, "rb") as f:
            data = f.read()

        # Find 'free' box
        free_idx = data.find(b'free')
        if free_idx == -1:
            return None

        # Check all known signatures
        for signature, parser_func in self.parsers.items():
            sig_idx = data.find(signature, free_idx)
            if sig_idx != -1:
                return parser_func(data, sig_idx)

        return None

    def _parse_v1_timestamp(self, data, rhom_idx):
        """Parse v1 timestamp format."""
        timestamp_bytes = data[rhom_idx + 4:rhom_idx + 12]
        timestamp_ms = int.from_bytes(timestamp_bytes, byteorder="big")

        return {
            'version': 1,
            'type': 'timestamp',
            'timestamp_ms': timestamp_ms,
            'datetime': datetime.fromtimestamp(timestamp_ms / 1000.0, tz=datetime.timezone.utc)
        }

    def add_parser(self, signature, parser_func):
        """Add custom parser for new metadata formats."""
        self.parsers[signature] = parser_func

# Usage
parser = RhombusMetadataParser()
metadata = parser.parse("video_segment.mp4")

if metadata:
    print(f"Version: {metadata['version']}")
    print(f"Type: {metadata['type']}")
    print(f"Timestamp: {metadata['datetime']}")

Performance considerations

Efficient File Reading

For large files, read only the header portion instead of the entire file

Caching Strategy

Cache extracted timestamps to avoid re-parsing the same files

Batch Processing

Process multiple files in parallel when building timelines

Memory Management

Use streaming parsers for very large video files

Optimized file reading

Optimized Parserpython
def extract_rhombus_timestamp_optimized(file_path, max_search_bytes=100_000):
    """
    Optimized version that reads only the beginning of the file.
    Most metadata is in the first portion of MP4 files.
    """
    with open(file_path, "rb") as f:
        # Read only header portion
        data = f.read(max_search_bytes)

    # Search for 'rhom' signature
    rhom_idx = data.find(b'rhom')
    if rhom_idx == -1:
        return None, None

    # Verify we have enough data for timestamp
    if len(data) < rhom_idx + 12:
        return None, None

    timestamp_bytes = data[rhom_idx + 4:rhom_idx + 12]
    timestamp_ms = int.from_bytes(timestamp_bytes, byteorder="big")
    timestamp_dt = datetime.fromtimestamp(timestamp_ms / 1000.0, tz=datetime.timezone.utc)

    return timestamp_ms, timestamp_dt

Summary

Each ISOBMFF segment carries a millisecond-precision UTC timestamp in its free atom. Standard video tools ignore the atom, so the segments play anywhere, and your code can read it to align events across streams to well under a second.

Tip

Next Steps for Developers:

  • Experiment with the ISOBMFF GitHub library to parse Rhombus segments
  • Use the MP4Box.js online viewer to visually inspect box structures
  • Incorporate timestamp extraction into your ingest pipeline for perfectly synchronized multi-source datasets

Additional resources

Support

Need assistance with timestamp extraction or video synchronization?