[Paper Review] BrnoCompSpeed: Review of Traffic Camera Calibration and Comprehensive Dataset for Monocular Speed Measurement.

Jakub Sochor, Roman Juránek+5 more|arXiv (Cornell University)|Feb 21, 2017

Autonomous Vehicle Technology and Safety16 citations

TL;DR

This paper introduces BrnoCompSpeed, a comprehensive monocular camera calibration and speed measurement dataset with 18 full-HD videos, 20,865 vehicle instances, and ground truth speeds from LiDAR and GPS. It evaluates a fully automatic calibration and speed estimation method, providing a benchmark for fair comparison of monocular speed measurement techniques.

ABSTRACT

In this paper, we focus on traffic camera calibration and a visual speed measurement from a single monocular camera, which is an important task of visual traffic surveillance. Existing methods addressing this problem are difficult to compare due to a lack of a common data set with reliable ground truth. Therefore, it is not clear how the methods compare in various aspects and what factors are affecting their performance. We captured a new data set of 18 full-HD videos, each around 1 hr long, captured at six different locations. Vehicles in the videos (20865 instances in total) are annotated with the precise speed measurements from optical gates using LiDAR and verified with several reference GPS tracks. We made the data set available for download and it contains the videos and metadata (calibration, lengths of features in image, annotations, and so on) for future comparison and evaluation. Camera calibration is the most crucial part of the speed measurement; therefore, we provide a brief overview of the methods and analyze a recently published method for fully automatic camera calibration and vehicle speed measurement and report the results on this data set in detail.

Motivation & Objective

• Address the lack of standardized datasets with reliable ground truth for monocular traffic speed measurement.
• Provide a comprehensive, publicly available dataset to enable fair comparison of existing and future speed measurement methods.
• Evaluate the performance of a fully automatic camera calibration and vehicle speed estimation pipeline on a real-world dataset.
• Analyze factors influencing the accuracy of monocular speed measurement in urban traffic surveillance.
• Establish a reproducible benchmark for future research in visual traffic surveillance using single-camera systems.

Proposed method

• Captured 18 full-HD video sequences, each approximately 1 hour long, from six distinct urban traffic locations.
• Collected precise ground truth speeds using optical gates synchronized with LiDAR and cross-verified with multiple GPS tracks.
• Annotated 20,865 vehicle instances with bounding boxes and speed values, ensuring high-precision labeling.
• Provided detailed camera calibration metadata, including intrinsic and extrinsic parameters, for each video.
• Implemented and evaluated a fully automatic camera calibration method using feature detection and geometric constraints.
• Integrated the calibration results into a monocular speed measurement pipeline based on 2D motion estimation and known road geometry.

Experimental results

Research questions

• RQ1 How does the performance of a fully automatic monocular camera calibration method vary across diverse urban traffic scenarios?
• RQ2 To what extent do errors in camera calibration propagate into speed measurement inaccuracies?
• RQ3 How consistent are speed estimates when validated against multiple ground truth sources (LiDAR and GPS)?
• RQ4 What factors—such as camera angle, vehicle speed, or road geometry—most significantly affect speed measurement accuracy?
• RQ5 Can a publicly available dataset with multi-source ground truth enable reliable benchmarking of monocular speed measurement algorithms?

Key findings

• The BrnoCompSpeed dataset contains 18 high-quality video sequences with 20,865 annotated vehicles and multi-source ground truth from LiDAR and GPS.
• The dataset enables precise evaluation of monocular speed measurement due to its high-accuracy ground truth and detailed calibration metadata.
• The fully automatic calibration method achieved consistent results across different camera configurations, demonstrating robustness in real-world settings.
• Speed measurement errors were significantly reduced when using the calibrated camera model compared to uncalibrated or approximated setups.
• The dataset revealed that camera angle and vehicle distance from the camera are key factors affecting speed estimation accuracy.
• The availability of the dataset allows for reproducible evaluation and fair comparison of future monocular speed measurement methods.