Roadside capture · side-fire · track

Side-fire video, organized around the vehicle track.

Tollscopic DOT# watches the vehicle pass from the side, tracks the physical vehicle through the camera view, selects evidence frames, and sends a durable package to the cloud.

01 · Why side-fire

Door markings are on the side. The camera needs to be too.

A plate camera watches the front or rear of the vehicle. For the carrier panel, the right geometry is side-fire — the camera sees the tractor door as the vehicle moves through the calibrated capture zone.

FIG. Top-down view. The vehicle moves through the side-fire capture zone; positions 3 and 4 produce the most useful door-panel angles and become selected evidence frames. Selection happens at the edge, across the track — not on the first frame the camera saw.

02 · Tracking and eligibility

Detect first. Track second. Gate third.

The edge detects vehicles, tracks each physical vehicle through the field of view, and uses calibrated geometry to admit likely commercial-vehicle candidates. Recall is prioritized — plausible commercial vehicles are not silently dropped.

Vehicle detection

Per-frame detection finds vehicles in the capture zone. No assumptions about lane or direction at this stage.

Physical tracking

Each detection is associated to a vehicle track. The track, not the frame, is the evidence unit.

Commercial eligibility

Calibration and geometry gate likely commercial vehicles. Downstream classes absorb false positives honestly.

03 · Frame selection

The best frame is the one that helps the model read.

The system looks across the vehicle track and selects frames with useful angle, sharpness, placard visibility, and temporal diversity. The first frame and the largest frame are not always the answer.

Too far

rejected

Motion blur

rejected

SELECTED

Good panel

kept

Glare

rejected

SELECTED

Good name

kept

Trailer occludes

rejected

Selection looks across the whole track. Sharpness, angle, panel visibility, and temporal diversity all matter. The system keeps the frames that are most likely to help the model read, not the frames that look best to a human eye.

04 · Durable track package

What leaves the roadside is a package, not a stream.

The edge buffers through network problems and uploads a single durable package per vehicle track. The cloud receives selected evidence frames with the metadata it needs to reason about them.

track_package · v1

{
  "track_id":     "tk_2026_05_20_a8c2f",
  "site_id":      "plaza_42",
  "lane":         "L3",
  "captured_at":  "2026-05-20T14:33:05Z",
  "vehicle_class_hint": "tractor+trailer",
  "selected_frames": [
    { "frame_id": "03", "uri": "...", "quality": 0.92 },
    { "frame_id": "04", "uri": "...", "quality": 0.88 }
  ],
  "trajectory_summary": "...",
  "calibration_ref":    "cal_l3_2026q2",
  "camera_id":          "cam_l2",
  "edge_version":       "edge-2026.05-a"
}

WHAT'S IN THE PACKAGE

Selected frame refs

evidence frames the cloud will reason over

Timing

captured_at + per-frame timestamps for replay

Camera + calibration

context the cloud needs for geometry

Trajectory summary

compact representation of the vehicle path

Track telemetry

edge confidence, detection counts, quality

Deployments can be configured so the cloud receives selected evidence frames and metadata rather than a continuous raw-video stream.

The edge keeps the track. The cloud does the rest.

Side-fire capture, calibrated zones, vehicle tracking, frame selection, and durable upload. The roadside part is meant to be practical.

Talk to Tollscopic →