Jello effect mitigation

Summary¶

Jello effect in drone video is caused by rolling shutter reading successive pixel rows at slightly different moments while the camera vibrates. The fix is to reduce vibration amplitude at the camera before it reaches the sensor. The mitigation hierarchy follows the same principle as all vibration control: eliminate at source first (prop balance), then isolate (floating motor mounts), then filter in software (gyroflow stabilisation). Each layer addresses a different frequency range. All three layers working together produce footage clean enough for professional use from a non-gimbal platform.

Concept¶

Rolling shutter and why jello appears¶

A global-shutter sensor reads all pixels at exactly the same instant — a perfect snapshot. A rolling-shutter sensor (used in nearly all small cameras and drones) reads pixel rows sequentially: row 1, then row 2, then row 3... all the way to the last row. The time to read a full frame might be 1/1000 s. If the camera is vibrating during that readout, row 500 is captured with the camera at a slightly different position than row 1.

The result: horizontal lines in the image are displaced relative to each other, creating the characteristic wobbly distortion. A stationary scene appears to wriggle; a straight propeller tip traces a wobbly arc.

At typical drone vibration frequencies (300–600 Hz), the readout time per row (approximately 15–30 µs for a typical 1080p sensor) captures the camera at meaningfully different positions on each row, producing visible jello.

The mitigation hierarchy¶

Layer 1 — Prop balance (eliminate at source)

An unbalanced propeller vibrates at its rotation frequency (300–600 Hz at operational RPM). This is the highest-amplitude single-frequency vibration source. Three minutes of magnetic prop balancing before the first flight of the day reduces this dramatically. Measure before and after with Blackbox — the noise floor change at motor fundamental frequency is visible.

Layer 2 — Floating motor mounts (mechanical isolation)

O-ring isolators between motor and arm provide approximately 10–20 dB of attenuation at motor vibration frequencies. This is the primary structural defence against jello. Worn or missing O-rings eliminate this protection. Inspect at every pre-flight. Replace on schedule (every 20–30 flight hours) or whenever cracked/flattened.

Layer 3 — Camera mounting (soft mount the camera)

If the camera is rigidly mounted to the airframe, all residual vibration that passes through the motor mounts reaches the camera directly. Mounting the camera on a soft pad (foam, silicone, or TPU printed dampers) adds a second mechanical isolation stage specifically at the camera. For fixed cameras without a gimbal, this is the highest-impact single change.

On libdrone's GPS/camera bracket, the HDZero camera mounts in a PETG slot. Adding a thin layer of anti-vibration foam between camera body and slot wall provides additional damping at minimal mass cost.

Layer 4 — Software stabilisation (remove residual)

Post-processing software (Gyroflow, DaVinci Resolve stabiliser) analyses the video and gyroscope data together to warp each frame, correcting for camera movement during the exposure. Gyroflow is particularly effective because it uses the actual IMU gyroscope data from Blackbox to compute the correction precisely rather than estimating it from the video content alone.

Software stabilisation crops the image (to allow warping room) and cannot fix jello caused by very high-frequency vibration (above the gyroscope's useful bandwidth). It is the final layer, not a substitute for the mechanical layers.

Reference¶

Jello mitigation effectiveness (approximate, cumulative)¶

These are representative values — actual results depend on specific platform, vibration frequency profile, and camera model.

Gyroflow integration with libdrone¶

Gyroflow works best with synchronised IMU data. To use Gyroflow with libdrone:

1. Enable Blackbox at maximum rate (blackbox_rate_denom = 1).
2. In Gyroflow: import the video file and the Blackbox log from the same flight.
3. Sync: align the gyro data with the video using motion matching or manual sync markers (a sharp movement visible in both the video and the gyro trace).
4. Apply correction. Gyroflow warps each frame based on the actual measured camera movement.

The Matek H7A3-SLIM's IMU (ICM-42688-P) is well-characterised in Gyroflow's lens/IMU database.

Procedure¶

Checking jello level after build changes¶

Any change to the motor mounts, motors, or propellers should be followed by a jello check:

1. Fly a slow (< 3 m/s) lateral pass over a flat, textured surface (grass, pavement, roof tiles).
2. Review the footage frame-by-frame. Look for horizontal waviness in what should be straight lines (pavement cracks, fence posts).
3. Check the Blackbox gyro spectrum simultaneously — peaks at motor RPM harmonics correlate with jello.
4. If jello has increased: check prop balance, check O-ring condition, check that the passive cover is not touching the arm head surface directly.

Rationale¶

Why all four layers are needed rather than relying on software alone¶

Software stabilisation is tempting as a complete solution — add Gyroflow and forget mechanical mitigation. This fails in practice for two reasons:

First, software stabilisation crops the image. On a narrow-FOV FPV camera, a 10–15% crop can eliminate context that is part of the shot. The more residual jello, the more aggressive the crop needed to remove it.

Second, very high-frequency jello (frame-to-frame row displacement at 600 Hz) introduces artefacts that software cannot remove because the gyroscope samples at 8 kHz but the correction is applied per-frame (at 30–60 Hz). Software stabilisation corrects frame-level motion, not within-frame row displacement. Mechanical mitigation is the only path to eliminating within-frame distortion.

Connections¶

requires: - aerial-imaging-basics - floating-motor-mounts - vibration-isolation-theory related: - pid-tuning-rate-profile - blackbox-analysis leads_to: - survey-imaging