Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Label Correction for Road Segmentation Using Road-side Cameras

Henrik Toikka, Eerik Alamikkotervo, Risto Ojala

TL;DR

The paper tackles the challenge of obtaining weather-robust road segmentation data by leveraging fixed roadside camera feeds to create a scalable labeling workflow. It introduces a semi-automatic label transfer pipeline that uses Fourier-Mellin image registration to propagate a single manual annotation across frames, augmented by an optimal-path strategy to handle large viewpoint changes. Validating on 927 Finland roadside cameras and testing on both roadside and dashcam data, the approach enables training of multiple segmentation models and demonstrates that corrected label reuse yields the best performance. The work offers a practical pathway to richer, weather-diverse perception data for autonomous driving and ADAS, while highlighting trade-offs between model backbone choice and labeling efficiency.

Abstract

Reliable road segmentation in all weather conditions is critical for intelligent transportation applications, autonomous vehicles and advanced driver's assistance systems. For robust performance, all weather conditions should be included in the training data of deep learning-based perception models. However, collecting and annotating such a dataset requires extensive resources. In this paper, existing roadside camera infrastructure is utilized for collecting road data in varying weather conditions automatically. Additionally, a novel semi-automatic annotation method for roadside cameras is proposed. For each camera, only one frame is labeled manually and then the label is transferred to other frames of that camera feed. The small camera movements between frames are compensated using frequency domain image registration. The proposed method is validated with roadside camera data collected from 927 cameras across Finland over 4 month time period during winter. Training on the semi-automatically labeled data boosted the segmentation performance of several deep learning segmentation models. Testing was carried out on two different datasets to evaluate the robustness of the resulting models. These datasets were an in-domain roadside camera dataset and out-of-domain dataset captured with a vehicle on-board camera.

Label Correction for Road Segmentation Using Road-side Cameras

TL;DR

The paper tackles the challenge of obtaining weather-robust road segmentation data by leveraging fixed roadside camera feeds to create a scalable labeling workflow. It introduces a semi-automatic label transfer pipeline that uses Fourier-Mellin image registration to propagate a single manual annotation across frames, augmented by an optimal-path strategy to handle large viewpoint changes. Validating on 927 Finland roadside cameras and testing on both roadside and dashcam data, the approach enables training of multiple segmentation models and demonstrates that corrected label reuse yields the best performance. The work offers a practical pathway to richer, weather-diverse perception data for autonomous driving and ADAS, while highlighting trade-offs between model backbone choice and labeling efficiency.

Abstract

Reliable road segmentation in all weather conditions is critical for intelligent transportation applications, autonomous vehicles and advanced driver's assistance systems. For robust performance, all weather conditions should be included in the training data of deep learning-based perception models. However, collecting and annotating such a dataset requires extensive resources. In this paper, existing roadside camera infrastructure is utilized for collecting road data in varying weather conditions automatically. Additionally, a novel semi-automatic annotation method for roadside cameras is proposed. For each camera, only one frame is labeled manually and then the label is transferred to other frames of that camera feed. The small camera movements between frames are compensated using frequency domain image registration. The proposed method is validated with roadside camera data collected from 927 cameras across Finland over 4 month time period during winter. Training on the semi-automatically labeled data boosted the segmentation performance of several deep learning segmentation models. Testing was carried out on two different datasets to evaluate the robustness of the resulting models. These datasets were an in-domain roadside camera dataset and out-of-domain dataset captured with a vehicle on-board camera.

Paper Structure

This paper contains 14 sections, 1 equation, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Roadside Cameras in Intelligent Transport Systems
  4. Image registration
  5. Road segmentation in challenging conditions
  6. Methods
  7. Image registration
  8. Finding optimal path of transformations
  9. Datasets
  10. Roadside camera dataset
  11. Dashcam dataset
  12. Validation
  13. Results
  14. Discussion

Figures (4)

  • Figure 1: Original manual labels are shown on the left column. Center and right columns display another frame from the same camera with a directly reused label and a corrected label, respectively.
  • Figure 2: Label correction architecture. Labels are aligned based on image registration, and cars possibly present on the road are segmented out of the label.
  • Figure 3: Visualisation of results on the Dashcam dataset.
  • Figure 4: Visualization of results on the roadside camera dataset.