Annotation-Free Curb Detection Leveraging Altitude Difference Image
Fulong Ma, Peng Hou, Yuxuan Liu, Ming Liu, and Jun Ma
HKUST(GZ) HKUST
[paper] [code]
This webpage is under construction.
Embedded Files
Motivation
Image-based curb detection methods are susceptible to fluctuations in lighting conditions, exhibiting poor robustness.
LiDAR-based curb detection methods often encounter significant processing delays due to the large number of 3D points contained in the point cloud data.
The unstructured nature of point cloud data poses challenges for integrating the latest advancements in deep learning into point cloud applications.
Manual data annotation is costly, time-consuming, and labor-intensive, making annotation-free curb detection highly significant.
The top row represents RGB images, while the bottom row represents ADIs. It can be observed that under these extreme lighting conditions, the curb in the RGB images is difficult to discern (as indicated by the red dashed boxes). However, the curb is clearly visible inthe ADIs.
Method
The overall framework of our proposed curb detection method based on ADIs with an annotation-free approach. (a): Schematic diagram of the network architecture. (b): Illustration of the process in the ACA module.
Mobileone module
Mobileone module
The MobileOne module has two different structures during trainingand inference. Top: MobileOne module during training with reparameterizable branches. Bottom: MobileOne module during inference where the branches are reparameterized.
Post-processing
Post-processing
Data flow throughout the entire process from inputting the ADI into the network to obtaining the final curb detection results.
Results
Comparison of our method with other methods on the KITTI 3D curb dataset.
Curb detection results of the same model when the input is RGB images instead of ADI.
Preprocessing, model inference, post-processing, and total time consumption.
Page updated
Google Sites
Report abuse