Road Detection

Road detection is one of the key issues in a smart vehicle developing, so a lot of research has been done. In the road features extraction based on images, it is possible to use computational vision techniques and digital image processing, more specifically, segmentation methods, thus separating the whole region considered as road from the rest of the image.

---

Road detection considering different patterns on the surface

Using Convolutional Neural Networks we can detect different types of road surfaces, such as: asphalt, unpaved roads or different types of pavement. We can also detect different patterns along the road, such as: potholes and water-puddles.

---

More details and information in: Original Paper

Step-by-step: Towards Data Science

---

Source code used: GitHub

---

---

@misc{rateke:2021,
  title = {Road surface detection and differentiation considering surface damages},
  author = {Thiago Rateke and Aldo von Wangenheim},
  journal = {Autonomous Robots},
  year = {2021},
  month = {Jan},
  day = {11},
  issn = {1573-7527},
  doi = {10.1007/s10514-020-09964-3},
  url = {https://doi.org/10.1007/s10514-020-09964-3}
}

---

Road Surface and Quality Classifcation

Our road type and quality classifier was done through a simple Convolutional Neural Network with few steps and presents promising results in different datasets.

---

More details and information in: ResearchGate

Step-by-step: Towards Data Science

---

Source code used: GitHub

---

* In this example we use dataset images from CaRINA.

---

@article{rtk:2019,
  author = {Thiago Rateke and Karla Aparecida Justen and Aldo von Wangenheim},
  title = {Road Surface Classification with Images Captured From Low-cost Cameras – Road Traversing Knowledge (RTK) Dataset},
  journal = {Revista de Informática Teórica e Aplicada (RITA)},
  year = {2019},
  doi = {https://doi.org/10.22456/2175-2745.91522},
}

---

Classical Computer Vision Road Detection Approach

Our work uses region-based techniques, not depending on the geometric models of the road (which could require a great computational cost for comparisons) and also not depending on lane markings (where often, due to lack of maintenance, there are no markings). More specifically, a region-based approach, where, based on pixels color, the image should be segmented into regions considered or not as road. In the current phase of the work segmentation methods are used (Canny Edge Detection, Flood Fill) based on pixels features (like color) to identify the path.

---

The process can be separate into three steps: pre-processing, image segmentation, and path location

Pre-processing

• Image color space transformation from RGB to CIE-Lab, which uses a specific channel for brightness.
• And then an histogram equalization is applied.
• Finally, an edge detection is performed, which must find all the possible borders in the image.

Image segmentation

After the pre-processing, the segmentation is started, which aims to divide the image into different homogeneous areas. The grouping of the pixels or sub-regions into larger regions is performed taking into account some input pixels, known as seed points, and from these points are grouped nearby and similar regions.

Path location

Finally, after the segmented image, the location of the navigable path occurs. The goal is to find which of the previously obtained areas is the road on the image. This step is performed considering that the navigable path will exist within a region of interest in the image, presented by a red rectangle.

Post-processing

• It consists of a morphological closing operation (dilatation followed by erosion).

---

* We used, in the experiments, images from the case base provided by KITTI.