Lane detection methods based on semantic segmentation

Image Segmentation is dividing objects in an image into meaningful units. To be more specific, it predicts which class each pixel in the image belongs to. When semantic segmentation is performed on an image, it is known which class each pixel belongs to, as shown in the following figure. Image segmentation is used in a wide range of fields such as medical image analysis (tumor boundary extraction, etc.), autonomous vehicles (road surface, pedestrian detection, etc.), and augmented reality. The main purpose of SegNet is a model designed to perform pixel-wise semantic segmentation of structures related to autonomous driving, such as roads, buildings, cars, and pedestrians. U-Net is mainly used in bio images, and instead of using Pooling Indices, the entire feature map is transferred from the encoder to the decoder and then concatenated to perform convolution. This makes the model larger and uses more memory.