Image analysis system, image analysis method, and image analysis program

1. An image analysis system having an image analysis server which analyzes an input image of a monitoring area and detects a state of a specific monitored object,

wherein the image analysis server is configured to

divide, in the input image of the monitoring area, a portion of an area in which the monitored object is expected to be present into a plurality of grids,

generate a trained model trained by associating “1” with an image of a grid in which the monitored object is present and “0” with an image of a grid in which the monitored object is not present for each grid,

calculate, for the input image, a confidence of a presence of the monitored object for each grid using the trained model,

determine a presence or absence of the monitored object in the monitoring area based on a comparison result between confidences of the plurality of grids and a threshold value, and

detect an abnormality when a confidence value in at least one of the grids is inconsistent with the determination result.

2. The image analysis system of claim 1 , wherein the image analysis server is configured to obtain information indicating the presence or absence of the monitored object in the monitoring area from an outside, and detect the abnormality when the confidence value in at least one of the grids is inconsistent with the information indicating the presence or absence of the monitored object.

3. The image analysis system of claim 2 , wherein the image analysis server is configured to accumulate a set of a grid unit image and information on the presence or absence of the monitored object corresponding to the grid unit image as training data, perform training using the training data to build a new model, and update the trained model with the new model.

4. The image analysis system of claim 3 , wherein the image analysis server is configured to exclude an image in which an abnormality is detected and information on the presence or absence of the monitored object corresponding to the image in which the abnormality is detected from the training data.

5. The image analysis system of claim 1 , wherein the image analysis server is configured to obtain information indicating the presence or absence of the monitored object in the monitoring area from an outside, and detect an abnormality when the information indicating the presence or absence of the monitored object obtained from the outside is different from the determination result.

6. The image analysis system of claim 5 , wherein the image analysis server is configured to accumulate a set of a grid unit image and information on the presence or absence of the monitored object corresponding to the grid unit image as training data, perform training using the training data to build a new model, and update the trained model with the new model.

7. The image analysis system of claim 6 , wherein the image analysis server is configured to exclude an image in which an abnormality is detected and information on the presence or absence of the monitored object corresponding to the image in which the abnormality is detected from the training data.

8. The image analysis system of claim 1 , wherein the image analysis server is configured to accumulate a set of a grid unit image and information on the presence or absence of the monitored object corresponding to the grid unit image as training data, perform training using the training data to build a new model, and update the trained model with the new model.

9. The image analysis system of claim 8 , wherein the image analysis server is configured to exclude an image in which an abnormality is detected and information on the presence or absence of the monitored object corresponding to the image in which the abnormality is detected from the training data.

10. An image analysis method for detecting a state of a specific monitored object by analyzing an input image of a monitoring area, comprising:

dividing, in the input image of the monitoring area, a portion of an area in which the monitored object is expected to be present into a plurality of grids,

generating a trained model trained by associating “1” with an image of a grid in which the monitored object is present and “0” with an image of a grid in which the monitored object is not present for each grid,

calculating, for the input image, a confidence of a presence of the monitored object for each grid using the trained model,

determining a presence or absence of the monitored object in the monitoring area based on a comparison result between confidences of the plurality of grids and a threshold value, and

detecting an abnormality when a confidence value in at least one of the grids is inconsistent with the determination result.

11. An image analysis program which operates on an image analysis server which is configured to analyze an input image of a monitoring area and detect a state of a specific monitored object,

wherein the image analysis program causes the image analysis server to perform dividing, in the input image of the monitoring area, a portion of an area in which the monitored object is expected to be present into a plurality of grids,

generating a trained model trained by associating “1” with an image of a grid in which the monitored object is present and “0” with an image of a grid in which the monitored object is not present for each grid,

calculating, for the input image, a confidence of a presence of the monitored object for each grid using the trained model,

obtaining information indicating a presence or absence of the monitored object in the monitoring area from an outside, and

detecting an abnormality when a confidence value in at least one of the grids is inconsistent with the information indicating the presence or absence of the monitored object,

wherein the presence or absence of the monitored object in the monitoring area is determined based on a comparison result between confidences of the plurality of grids and a threshold value.