Road Curb Detection and Localization with Monocular Forward-view Vehicle Camera (2002.12492v1)

Abstract: We propose a robust method for estimating road curb 3D parameters (size, location, orientation) using a calibrated monocular camera equipped with a fisheye lens. Automatic curb detection and localization is particularly important in the context of Advanced Driver Assistance System (ADAS), i.e. to prevent possible collision and damage of the vehicle's bumper during perpendicular and diagonal parking maneuvers. Combining 3D geometric reasoning with advanced vision-based detection methods, our approach is able to estimate the vehicle to curb distance in real time with mean accuracy of more than 90%, as well as its orientation, height and depth. Our approach consists of two distinct components - curb detection in each individual video frame and temporal analysis. The first part comprises of sophisticated curb edges extraction and parametrized 3D curb template fitting. Using a few assumptions regarding the real world geometry, we can thus retrieve the curb's height and its relative position w.r.t. the moving vehicle on which the camera is mounted. Support Vector Machine (SVM) classifier fed with Histograms of Oriented Gradients (HOG) is used for appearance-based filtering out outliers. In the second part, the detected curb regions are tracked in the temporal domain, so as to perform a second pass of false positives rejection. We have validated our approach on a newly collected database of 11 videos under different conditions. We have used point-wise LIDAR measurements and manual exhaustive labels as a ground truth.

• The paper introduces a real-time curb detection system that leverages monocular vision, fisheye rectification, and edge extraction for precise localization.
• It employs 3D template fitting with SVM classification on HOG features to achieve over 90% detection accuracy and mean distance errors below 9%.
• Temporal tracking using autoregressive models ensures smooth curb parameter transitions, enhancing the reliability and cost-effectiveness of ADAS.

Overview of Road Curb Detection and Localization Using Monocular Camera

The paper "Road Curb Detection and Localization with Monocular Forward-view Vehicle Camera" presents a real-time curb detection and localization method using a monocular camera equipped with a fisheye lens. This method is particularly relevant in the context of Advanced Driver Assistance Systems (ADAS), which are essential for ensuring vehicle safety during parking maneuvers. The approach combines computer vision techniques with geometric reasoning to estimate the distance, orientation, height, and depth of road curbs relative to a vehicle.

Key Methodology

The research outlines a two-component system comprising curb detection in individual video frames and subsequent temporal analysis. For curb detection, the method involves:

1. Curb Edge Extraction: Using Canny edge detection and Hough transform to identify line segments corresponding to curb edges in each frame after rectifying distortions from the fisheye lens.
2. 3D Template Fitting: A parametric 3D curb template is fitted to the detected edge lines. The system uses the geometric properties of curbs, assuming straight edges and simple prismatic shapes.
3. Outlier Rejection: Non-curb features are filtered out using Support Vector Machine (SVM) classifiers trained on Histograms of Oriented Gradients (HOG) features.
4. Temporal Tracking: The approach leverages temporal continuity ensuring smooth transitions in the curb parameters across frames, implementing an autoregressive model to predict the curb's position based on previous frames.

Results and Validation

The system's efficacy was tested using a database of 11 videos recorded under various conditions. The results indicated an impressive detection accuracy of over 90% for curb incidence across sequences. The mean error in estimating the curb-to-vehicle distance was consistently below 8-9% across different conditions and video sequences.

Challenges such as varying curb appearances under different environmental conditions, sensor noise, and occlusions were addressed, highlighting the robustness of the approach. The requirement of a single monocular camera, as opposed to more expensive LIDAR or stereo-vision systems, emphasizes its cost-effectiveness in ADAS applications.

Implications and Future Work

The research delineates significant implications for affordable ADAS solutions by leveraging monocular vision for environment perception. Accurate curb detection and localization enhance vehicle safety systems, such as automatic parking assistants, where precise environmental interaction is crucial.

Further developments could explore extensions to include various curb types and environmental conditions, such as inclement weather or nighttime scenarios, expanding the applicability of the method. Moreover, integrating deep learning approaches could enhance feature detection and classification in complex scenes, potentially increasing robustness and reliability.

In conclusion, the paper contributes to the field of autonomous and semi-autonomous vehicle navigation by proposing a method that efficiently utilized monocular vision systems for a critical component of vehicle-environment interaction.