Road extraction device and method
A road extraction device and method comprising: a data collection module configured to collect a road image associated with a road; and a deep learning module configured to perform a task for extracting the road from the road image by using deep learning technology, wherein the deep learning module comprises: a first deep learning unit configured to perform segmentation on the road image; and a second deep learning unit configured to perform object detection on the road image.
1 . A road extraction device comprising:
a data collection module configured to collect a road image associated with a road; and
a deep learning module configured to perform a task for extracting the road from the road image by using deep learning technology,
wherein the deep learning module comprises:
a first deep learning unit configured to perform segmentation on the road image; and
a second deep learning unit configured to perform object detection on the road image,
wherein the deep learning module is trained through multi-task learning,
wherein the deep learning module is configured to perform the task by using an autoencoder,
wherein the first deep learning unit and the second deep learning unit each comprise at least one encoding unit configured to convert the road image into a latent representation by encoding the road image, which is input data, and at least one decoding unit configured to generate output data by converting the latent representation,
wherein at least one of the first deep learning unit and the second deep learning unit comprises a plurality of decoding units,
wherein each of the plurality of decoding units is pre-trained to perform different types of tasks, wherein a 1st-1 decoding unit included in the first deep learning unit is configured to perform a task associated with at least one of road surface, road center line, texture representing a texture of the road, road edge representing an edge of the road, and road angle (direction) representing an angle of the road,
a 1st-2 decoding unit included in the first deep learning unit is configured to perform a task associated with a not-road mark including at least one of a building,
a plant, and water, and
a second decoding unit included in the second deep learning unit is configured to perform a task associated with a road mark including a car.
2 . The road extraction device of claim 1 , wherein the road image comprises a satellite image associated with the road.
3 . The road extraction device of claim 1 , further comprising a post-processing module configured to generate final data by merging the tasks performed by the first deep learning unit and the second deep learning unit and then processing the merged tasks.
4 . The road extraction device of claim 3 , wherein, when an unconnected area associated with the road exists, the post-processing module is configured to convert the unconnected area into a connected area.