中文Start a project

AVM ENGINEERING

Surround-view lens distortion: design for calibration, stitching and perception

DIRECT ANSWER

For surround-view cameras, “low distortion” is usually not the right universal target. The lens should provide a defined projection and repeatable distortion curve that preserves useful pixel density, supports calibration and stitching, and remains stable through temperature and production variation.

1. Separate FOV, projection and distortion

FOV describes angular coverage. Projection describes how scene angle maps to image height. Distortion describes deviation from a chosen reference model. A fisheye lens can intentionally use a non-rectilinear projection while still being well controlled.

State the intended projection or calibration model instead of requesting only a maximum distortion percentage.

  • Horizontal, vertical and diagonal FOV
  • Reference projection model
  • Allowed mapping residual
  • Monotonic angle-to-pixel behavior

2. Allocate pixels where the system needs them

Ultra-wide optics compress large scene angles into a finite sensor. The distortion strategy determines how many pixels represent objects near the vehicle, at the horizon and in overlap zones.

Define important regions with the perception and stitching teams. A curve that looks visually natural may not be the curve that best supports detection or seam quality.

  • Near-field ground coverage
  • Horizon and object-recognition region
  • Camera overlap zones
  • Pixels per angular interval

3. Protect peripheral image quality

AVM images depend heavily on the outer field, where aberrations, relative illumination and color shading are hardest to control. MTF should be reviewed at field points connected to stitching and detection, not only at the center.

Edge sharpness, lateral color and relative illumination interact with ISP correction. Agree on what is corrected optically, what is corrected digitally and what variation the algorithm can accept.

  • Peripheral sagittal / tangential MTF
  • Relative illumination at useful field
  • Lateral color and color shading
  • Optical versus ISP correction budget

4. Specify calibration repeatability

A nominal distortion curve is not enough if production lenses vary in mapping, optical center, tilt or effective focal behavior. These variations can change calibration residuals and seam positions.

Link lens tolerances and camera assembly tolerances to calibration limits. Include temperature and mechanical stress when they can move the mapping.

  • Distortion-curve repeatability
  • Optical center and image tilt
  • Camera pose and assembly variation
  • Temperature-dependent calibration residual

5. Evaluate the complete surround-view chain

The lens, cover glass, sensor, ISP, calibration model and stitching algorithm form one imaging system. Review representative images and quantitative mapping data together.

Before SOP, validate multiple cameras and production-representative units in day, night, wet, high-contrast and environmental conditions relevant to the vehicle location.

  • Lens-to-algorithm mapping data
  • Multi-camera overlap and seam review
  • Day / night / wet-scene images
  • Production-representative validation

FAQ

Frequently asked engineering questions

Should a surround-view fisheye lens have zero distortion?

Usually no. Very wide coverage requires a defined non-rectilinear projection. The goal is controlled, repeatable mapping that supports the intended calibration and algorithm.

Why is DFOV alone insufficient for AVM lens selection?

DFOV does not show horizontal and vertical coverage, pixel distribution, distortion curve, edge quality or overlap behavior.

Can distortion be corrected entirely in software?

Software can remap a known curve, but it cannot recover missing edge resolution, signal-to-noise ratio or unstable unit-to-unit mapping. Optical and digital budgets must be coordinated.

CONTINUE READING

Explore the automotive optics knowledge center

All engineering guides →