[Paper Review] Autocamera Calibration for traffic surveillance cameras with wide angle lenses

Aman Gajendra Jain, Nicolas Saunier|PolyPublie (École Polytechnique de Montréal)|Jan 20, 2020

Video Surveillance and Tracking Methods10 references4 citations

TL;DR

This paper presents an automatic camera calibration method for traffic surveillance cameras with wide-angle lenses using only video footage and camera height as input. It performs intrinsic calibration via equidistant fisheye model and extrinsic calibration using vanishing points from vehicle trajectories at perpendicular intersections, achieving ~6.35% average error in focal length estimation and superior undistortion performance compared to checkerboard calibration.

ABSTRACT

We propose a method for automatic calibration of a traffic surveillance camera with wide-angle lenses. Video footage of a few minutes is sufficient for the entire calibration process to take place. This method takes in the height of the camera from the ground plane as the only user input to overcome the scale ambiguity. The calibration is performed in two stages, 1. Intrinsic Calibration 2. Extrinsic Calibration. Intrinsic calibration is achieved by assuming an equidistant fisheye distortion and an ideal camera model. Extrinsic calibration is accomplished by estimating the two vanishing points, on the ground plane, from the motion of vehicles at perpendicular intersections. The first stage of intrinsic calibration is also valid for thermal cameras. Experiments have been conducted to demonstrate the effectiveness of this approach on visible as well as thermal cameras. Index Terms: fish-eye, calibration, thermal camera, intelligent transportation systems, vanishing points

Motivation & Objective

• To enable automatic calibration of traffic surveillance cameras with wide-angle lenses without manual scene annotations or 3D calibration rigs.
• To resolve scale ambiguity in extrinsic calibration using only the camera height as user input.
• To develop a method robust to lens distortion and independent of scene-specific features or prior calibration data.
• To extend applicability to thermal cameras by leveraging motion-based calibration.
• To enable downstream applications such as speed monitoring, 3D reconstruction, and vehicle tracking.

Proposed method

• Intrinsic calibration is performed using a polynomial radial distortion model with equidistant fisheye projection, estimating distortion coefficients from vehicle trajectory motion.
• The method assumes a pinhole camera model with square pixels, zero skew, and optical center at image center, simplifying the intrinsic matrix.
• Extrinsic calibration estimates two orthogonal vanishing points on the ground plane from the motion of vehicles at perpendicular intersections.
• Vanishing points are detected via RANSAC-based voting in the Hough space of trajectory directions, with robustness improved by averaging top 20% votes and using standard deviation.
• Rotation matrix is refined using SVD to enforce orthogonality, and translation vector is computed using the camera height to resolve scale ambiguity.
• The entire process runs on a few minutes of video footage, requiring only camera height as user input.

Experimental results

Research questions

• RQ1 Can intrinsic calibration of wide-angle cameras be achieved automatically using only moving vehicles and no scene-specific annotations?
• RQ2 How accurately can vanishing points be estimated from vehicle trajectories in a traffic scene?
• RQ3 To what extent does the proposed method reduce distortion compared to traditional checkerboard calibration?
• RQ4 Can the method be applied to thermal cameras with the same performance as visible-spectrum cameras?
• RQ5 Is the calibration robust to occlusions and non-straight-line vehicle motion?

Key findings

• The proposed method achieves an average focal length estimation error of 6.35% relative to checkerboard calibration, demonstrating strong accuracy in intrinsic calibration.
• The mean least square error (LSE) of trajectory undistortion was reduced to 0.006–0.007 after calibration, outperforming checkerboard-calibrated coefficients in 4 out of 5 datasets.
• The method produced visually accurate top-view transformations, with parallel road markings, circular manholes, and rectangular crosswalks, confirming correct extrinsic calibration.
• The algorithm successfully calibrated thermal cameras using the same motion-based approach, proving its applicability beyond visible-spectrum cameras.
• The method is robust to noise and performs well even with limited vehicle density, provided sufficient vehicles move in perpendicular directions.
• The approach is fully automatic and requires only a few minutes of video and camera height, making it practical for real-world deployment.