Periphery monitoring device
A periphery monitoring device includes: an acquisition unit that acquires capture images which are captured by an imaging unit in time series when a vehicle is moving, the imaging unit being provided in the vehicle and capable of imaging surroundings of the vehicle; and a restoration processing unit that, in a case where dirt is present in a latest capture image among the capture images, inputs the capture images into a first learned model and generates a first restoration image as a restoration image obtained by restoring a region concealed with the dirt in the latest capture image, the first learned model being a result obtained by machine learning a relationship between a first learning image in which learning dirt is not present and first learning dirt images each of which is made by causing the learning dirt to be present in the first learning image.
1. A periphery monitoring device comprising:
at least one processor configured to implement:
an acquisition unit that acquires a plurality of capture images which are captured by an imaging unit in time series when a vehicle is moving, the imaging unit being provided in the vehicle and capable of imaging surroundings of the vehicle; and
a restoration processing unit that, in a case where dirt is present in a latest capture image among the plurality of capture images, inputs the plurality of capture images into a first learned model and generates a first restoration image as a restoration image obtained by restoring a region concealed with the dirt in the latest capture image, the first learned model being a result obtained by machine learning a relationship between a first learning image in which learning dirt is not present and a plurality of first learning dirt images each of which is made by causing the learning dirt to be present in the first learning image.
2. The periphery monitoring device according to claim 1 , wherein
the restoration processing unit inputs the first restoration image and the latest capture image into a second learned model and restores a second restoration image as the restoration image, the second learned model being a result obtained by machine learning a relationship between a restoration image generated by inputting a plurality of second learning dirt images in each of which the learning dirt is present into the first learned model and a latest learning dirt image among the plurality of second learning dirt images used when restoring the restoration image.
3. The periphery monitoring device according to claim 2 , wherein
the restoration processing unit acquires dirt information indicating a size of the dirt present in the latest capture image, and in a case where the size of the dirt indicated by the dirt information is equal to or greater than a predetermined threshold, does not execute restoration processing.
4. The periphery monitoring device according to claim 3 , wherein
the restoration processing unit acquires speed information of the vehicle at the time when the acquisition unit acquires the capture image, and changes the threshold according to a vehicle speed indicated by the speed information.
5. The periphery monitoring device according to claim 1 , wherein
the restoration processing unit acquires dirt information indicating a size of the dirt present in the latest capture image, and in a case where the size of the dirt indicated by the dirt information is equal to or greater than a predetermined threshold, does not execute restoration processing.
6. The periphery monitoring device according to claim 5 , wherein
the restoration processing unit acquires speed information of the vehicle at the time when the acquisition unit acquires the capture image, and changes the threshold according to a vehicle speed indicated by the speed information.