サブフレーム精度
選手がフレームキャプチャのまさにその瞬間にゲートを通過することは稀です。当システムは複数のフレームを解析し、フレーム間の通過瞬間を特定します — カメラのネイティブフレーム間隔をはるかに超えるサブミリ秒の補間精度を実現します。
The interpolation works by observing the athlete's torso position in at least 3 consecutive frames as they approach and cross the timing line. By fitting a linear motion model to these positions, we solve for the precise timestamp when the torso center crossed the gate position. Because human sprinting velocity is nearly constant over short distances (the 1-2 meter detection zone), the linear model produces highly accurate results.
物理ベースの補正
phoneのカメラは各フレームを上から下へ数ミリ秒かけてスキャンするため、選手がフレーム内のどの位置に映るかによってタイミングオフセットが生じます。このほか、ハードウェアレベルのタイミングアーティファクトについても自動的に補正し、系統的誤差を排除します。
TrackSpeed automatically measures the rolling shutter duration for each device and applies a per-pixel timing correction. If an athlete's torso is detected at 40% of the frame height and the rolling shutter scan takes 10ms, we apply a 4ms correction to the timestamp. This correction is applied before interpolation, ensuring the trajectory regression operates on accurate timing data. The result: rolling shutter becomes a non-factor in the final timing.
マルチデバイスクロック同期
スタートに1台、フィニッシュに1台の2台のphoneを使用する場合、両者のクロックはミリ秒以内で一致する必要があります。当社の同期プロトコルはデバイス間で±3-5msの精度を実現します。同期中は、最良の結果を得るために両方のスマートフォンを約10m以内に置いてください。同期完了後は、必要な距離まで離してご使用いただけます。
The sync protocol sends multiple round-trip timing probes between the two phones via peer-to-peer WiFi. By measuring send and receive timestamps on both devices, we calculate the clock offset using the same mathematical principles as the Network Time Protocol (NTP). Multiple samples are taken and outliers are rejected, achieving ±3-5ms alignment between devices. During sync, keep both phones within about 10 meters of each other for best results. Once synced, move them as far apart as needed — the offset stays locked.
ビームブレイクではなくボディトラッキング
レーザーゲートはビームを最初に遮断したものに反応します — 伸ばした腕、膝、あるいは胸のいずれか。これはランニングフォームによって異なり、不一致の原因となります。TrackSpeedは選手の体の重心を追跡して胴体がラインを通過したタイミングを判定し、公式タイミングの方式と一致させます。
TrackSpeed's detection engine replicates the official approach. Our geometry-based algorithm processes each video frame at 180×320 resolution, using frame differencing to detect motion, connected component labeling to identify body regions, and size/shape filters to isolate the torso from limbs. An arm-spike suppression algorithm prevents raised arms from triggering false crossings. The result: consistent chest-based timing that matches the official standard, unlike laser gates that trigger on the nearest body part.
サブフレーム補間、物理補正、クロック同期の組み合わせにより、専用ハードウェアに匹敵するタイミング精度を実現します — phoneだけで。
他システムとの比較
| システム | 分解能 | 検知方式 | コスト |
|---|---|---|---|
| FAT（公式） | 0.001s | フォトフィニッシュカメラ | $10,000+ |
| レーザーゲート | ±50-200ms* | 最初のビームブレイク | $500-2000 |
| TrackSpeed | ~4ms | ボディトラッキング | お使いのphone |
The Detection Pipeline
Every video frame captured by your phone's camera goes through a multi-stage detection pipeline. Here's what happens in the roughly 2ms it takes to process each frame:
1. Downsample & Extract Luma
The raw camera frame is downsampled to 180×320 pixels and converted to a grayscale (luma) image. This resolution is sufficient for accurate body detection while keeping processing fast enough for real-time analysis at 120fps.
2. Frame Differencing
Each frame is compared against the previous frame (N-1 differencing). Pixels that changed significantly — above a fixed threshold of 15 — indicate motion. This produces a binary motion mask showing where the athlete is moving.
3. Connected Component Labeling
The motion mask is analyzed using an 8-connected component labeling algorithm, which groups adjacent motion pixels into distinct blobs. Each blob is a candidate body region.
4. Size & Shape Filters
Candidate blobs are filtered by physical plausibility. A valid body region must be at least 33% of the frame height and 8% of the frame width. Fill ratio (minimum 0.20) and aspect ratio (maximum 1.2) filters reject noise, shadows, and camera shake artifacts.
5. Gate Crossing Detection
When a validated body region crosses the timing line position, a crossing event is triggered. A 0.5-second cooldown prevents double-counting from the same athlete, and gate-occupied tracking ensures each crossing is recorded exactly once.
6. Position-Based Interpolation
The crossing timestamp is refined using trajectory regression across the surrounding frames. This is where the sub-frame precision comes from — converting a discrete frame event into a continuous-time measurement with ~4ms accuracy.
Device Stability & Thermal Management
Accurate detection requires a stable camera. TrackSpeed monitors your phone's gyroscope in real-time, alerting you if the device moves during timing. The IMU (Inertial Measurement Unit) detects rotation rates as small as 0.15 rad/s — about the level of vibration from a passing truck.
Running the camera at 120fps is computationally intensive. TrackSpeed continuously monitors the device's thermal state and will warn you if the phone is overheating, which can cause frame drops and reduce timing accuracy. For sessions longer than 30 minutes, we recommend using 60fps, which maintains ~6ms accuracy with significantly lower thermal impact.
Frequently Asked Questions
How accurate is phone-based sprint timing compared to laser gates?
TrackSpeed achieves ~4ms effective timing accuracy, which is more consistent than laser gates for training purposes. Laser gates have sub-millisecond trigger speed, but they trigger on whichever body part breaks the beam first — hand, knee, or chest — creating 50-200ms of run-to-run variability. TrackSpeed tracks the chest crossing, matching official timing standards and eliminating this inconsistency.
Does TrackSpeed work without internet or WiFi?
Yes. Single-phone timing works completely offline. Multi-phone timing uses peer-to-peer WiFi Direct between the devices — no internet connection or router needed. The phones connect directly to each other.
Can TrackSpeed replace professional timing equipment?
For training purposes, yes. TrackSpeed provides the consistency and accuracy needed to track progress, compare training sessions, and make coaching decisions. For official competition results that go on record, Fully Automatic Timing (FAT) systems with photo finish cameras remain the standard — they achieve 0.001s resolution. TrackSpeed is designed for the 99% of timing that happens in training, not the 1% that happens at meets.
Why does frame rate matter for timing accuracy?
Higher frame rates give more data points for trajectory regression. At 30fps, frames are 33.3ms apart; at 60fps, 16.7ms; at 120fps, 8.3ms. More frames near the crossing moment means better interpolation accuracy. However, even at 30fps, sub-frame interpolation achieves ~10ms accuracy — still far better than manual stopwatch timing (~200ms human reaction time error).
How does multi-phone timing work?
Place one phone at the start line and one at the finish. The phones connect via peer-to-peer WiFi and synchronize their clocks using an NTP-style protocol (±3-5ms accuracy). Each phone independently detects when the athlete crosses its line. The finish phone subtracts the start timestamp from the finish timestamp to calculate the elapsed time, corrected for the measured clock offset.
What happens if my phone overheats during a session?
TrackSpeed monitors thermal state and warns you before accuracy degrades. At 120fps, sustained recording can cause thermal throttling after 20-30 minutes, leading to dropped frames. If this happens, the app suggests switching to 60fps, which uses significantly less power while maintaining ~6ms timing accuracy. For long practice sessions, 60fps is recommended from the start.
トレーニングの計時を始めませんか？
TrackSpeedをダウンロードして、phoneだけでミリ秒精度を体験しましょう。
