Method and apparatus for training an object identification neural network, and computer device
A method for training an object identification neural network based on a distorted fisheye image includes: acquiring the distorted fisheye image, wherein the distorted fisheye image comprises at least one target object; dividing the distorted fisheye image into a plurality of zones according to a distortion degree; determining a zone to which each of the at least one target object belongs; categorizing the at least one target object according to the zone to which each of the at least one target object belongs; and inputting the distorted fisheye image and a category to which each of the at least one target object belongs into the object identification neural network to train the object identification neural network.
1. A method for training an object identification neural network based on a distorted fisheye image, comprising:
acquiring the distorted fisheye image, wherein the distorted fisheye image comprises at least one target object;
dividing the distorted fisheye image into a plurality of zones according to a distortion degree;
determining a zone to which each of the at least one target object belongs;
categorizing the at least one target object according to the zone to which each of the at least one target object belongs; and
inputting the distorted fisheye image and a category to which each of the at least one target object belongs into the object identification neural network to train the object identification neural network.
2. The method according to claim 1 , wherein the step of categorizing the at least one target object according to the zone to which each of the at least one target object belongs comprises:
placing identical target objects belonging to a same zone into a category.
3. The method according to claim 1 , wherein target objects in a same zone have the same distortion degree.
4. A method for training an object identification neural network based on a distorted fisheye image, comprising:
acquiring the distorted fisheye image, wherein the distorted fisheye image comprises at least one target object;
dividing the distorted fisheye image into a plurality of zones according to a distortion degree;
determining a zone to which each of the at least one target object belongs;
categorizing the at least one target object according to the zone to which each of the at least one target object belongs; and
inputting the distorted fisheye image and a category to which each of the at least one target object belongs into the object identification neural network to train the object identification neural network;
wherein the step of determining the zone to which each of the at least one belongs comprises:
defining a circumscribed block for each of the at least one target object in the distorted fisheye image;
calculating an overlap area between the circumscribed block of each of the at least one target object and each zone;
acquiring a maximum overlap zone corresponding to each of the at least one target object, wherein the maximum overlap zone is a zone whose overlap area between the circumscribed block of each of the at least one target object and each area is the maximum;
judging whether a quantity of the maximum overlap zone corresponding to each of the at least one target object is greater than 1;
taking the maximum overlap zone as the zone to which each of the at least one target object belongs if the quantity of the maximum overlap zone corresponding to each of the at least one target object is 1; and
if the quantity of the maximum overlap zone corresponding to each of the at least one target object is greater than 1, selecting an optimal zone from the maximum overlap zones, and taking the optimal zone as the zone to which each of the at least one target object belongs, wherein the optimal zone is a zone most proximal to a center of the distorted fisheye image among the maximum overlap zones.
5. The method according to claim 4 , wherein the step of categorizing the at least one target object according to the zone to which each of the at least one target object belongs comprises:
placing identical target objects belonging to a same zone into a category.
6. The method according to claim 4 , wherein target objects in a same zone have the same distortion degree.
7. A computing device, comprising: a processor, a memory, a communication interface and a communication bus; wherein
the processor, the memory and the communication interface communicate with each other via the communication bus; and
the memory is configured to store at least one executable instruction, wherein the executable instruction, when being executed by the processor, causes the processor to perform the steps of:
acquiring the distorted fisheye image, wherein the distorted fisheye image comprises at least one target object;
dividing the distorted fisheye image into a plurality of zones according to a distortion degree;
determining a zone to which each of the at least one target object belongs;
categorizing the at least one target object according to the zone to which each of the at least one target object belongs; and
inputting the distorted fisheye image and a category to which each of the at least one target object belongs into the object identification neural network to train the object identification neural network.
8. The computer device according to claim 7 , wherein the step of determining the zone to which each of the at least one belongs comprises:
defining a circumscribed block for each of the at least one target object in the distorted fisheye image;
calculating an overlap area between the circumscribed block of each of the at least one target object and each zone;
acquiring a maximum overlap zone corresponding to each of the at least one target object, wherein the maximum overlap zone is a zone whose overlap area between the circumscribed block of each of the at least one target object and each area is the maximum;
judging whether a quantity of the maximum overlap zone corresponding to each of the at least one target object is greater than 1;
taking the maximum overlap zone as the zone to which each of the at least one target object belongs if the quantity of the maximum overlap zone corresponding to each of the at least one target object is 1; and
if the quantity of the maximum overlap zone corresponding to each of the at least one target object is greater than 1, selecting an optimal zone from the maximum overlap zones, and taking the optimal zone as the zone to which each of the at least one target object belongs, wherein the optimal zone is a zone most proximal to a center of the distorted fisheye image among the maximum overlap zones.
9. The computer device according to claim 7 , wherein the step of categorizing the at least one target object according to the zone to which each of the at least one target object belongs comprises:
placing identical target objects belonging to a same zone into a category.
10. The computer according to claim 7 , wherein target objects in a same zone have the same distortion degree.
11. A non-transitory computer-readable storage medium, the storage medium storing at least one executable instruction; wherein the executable instruction, when being executed, causes a processor to perform the steps of:
acquiring the distorted fisheye image, wherein the distorted fisheye image comprises:
dividing the distorted fisheye image into a plurality of zones according to a distortion degree;
determining a zone to which each of the at least one target object belongs;
categorizing the at least one target object according to the zone to which each of the at least one target object belongs; and
inputting the distorted fisheye image and a category to which each of the at least one target object belongs into the object identification neural network to train the object identification neural network.
12. The non-transitory computer-readable storage medium according to claim 11 , wherein the step of determining the zone to which each of the at least one belongs comprises:
defining a circumscribed block for each of the at least one target object in the distorted fisheye image;
calculating an overlap area between the circumscribed block of each of the at least one target object and each zone;
acquiring a maximum overlap zone corresponding to each of the at least one target object, wherein the maximum overlap zone is a zone whose overlap area between the circumscribed block of each of the at least one target object and each area is the maximum;
judging whether a quantity of the maximum overlap zone corresponding to each of the at least one target object is greater than 1;
taking the maximum overlap zone as the zone to which each of the at least one target object belongs if the quantity of the maximum overlap zone corresponding to each of the at least one target object is 1; and
if the quantity of the maximum overlap zone corresponding to each of the at least one target object is greater than 1, selecting an optimal zone from the maximum overlap zones, and taking the optimal zone as the zone to which each of the at least one target object belongs, wherein the optimal zone is a zone most proximal to a center of the distorted fisheye image among the maximum overlap zones.
13. The non-transitory computer-readable storage medium according to claim 11 , wherein the step of categorizing the at least one target object according to the zone to which each of the at least one target object belongs comprises:
placing identical target objects belonging to a same zone into a category.
14. The non-transitory computer-readable storage medium according to claim 11 , wherein target objects in a same zone have the same distortion degree.