Method and system for detecting change to structure by using drone
Disclosed herein is a method for an image analysis server to detect a change to a structure by using a drone. The method for an image analysis server to detect a change to a structure by using a drone includes: receiving images of a specific inspection target structure taken at different time points by a drone; detecting the difference between an image taken at a first time point and an image taken at a second time point based on the received images; and detecting a change to the inspection target structure via the detected difference, and generating a risk signal and then transmitting it to an administrator terminal.
1. A method for an image analysis server to detect a change to a structure by using a drone, the method comprising:
receiving a first image of the structure taken at a first time and a second image of the structure taken at a second time by a drone;
acquiring a first feature map for the first image and a second feature map for the second image using a machine learning algorithm;
detecting a difference between a first feature of the first feature map and a second feature of the second feature map;
predicting a risk of the structure based on the detected difference using a recurrent neural network; and
measuring an area of an actual structure in which the change is detected based on the first feature map and the second feature map,
wherein detecting the difference comprises,
determining a relative size of the area of the structure in which the change is detected in the second feature map with a same angle view of the first feature map with respect to a total area of the structure in the first feature map, and
determining a risk level of the structure by comparing the area of the structure where the change is detected with a preset value,
wherein measuring the area of the actual structure comprises,
calculating a ratio between a first distance between feature points in the second feature map of the structure and a second distance between feature points in the actual structure, and
calculating the area of the actual structure by applying the ratio to a change region within the second feature map of the structure,
wherein the recurrent neural network comprises a long short-term memory algorithm generating a new image capable of predicting the change of the structure over time.
2. The method of claim 1 , wherein the second time comprises a plurality of time points, and the second image of the structure comprises a plurality of images taken corresponding to each of the plurality of the time points,
wherein the new image is generated based on the plurality of images using the long short-term memory algorithm.
3. The method of claim 2 , wherein an interval of time between the plurality of the time points is constant.
4. The method of claim 1 , wherein the risk level comprises a first risk level defined when a change region value per unit time in the area of the structure is greater than or equal to a first preset value and less than a preset second value, and a second risk level defined when the change region value per unit time in the area of the structure is greater than or equal to the preset second value and less than a third preset value.