Swimmer monitoring device and swimmer monitoring method

Abstract

To provide a swimmer monitoring device and swimmer monitoring method that prevent drowning accidents in swimming areas without burdening swimmers. [Solution] The swimmer monitoring method includes means for acquiring monitoring images of a swimming area taken from above the water surface, means for acquiring swimming state information output from a trained model by inputting the acquired monitoring images into an AI model, and means for generating swimmer drowning warning information based on the acquired swimming state information. The swimming state information includes detection information of swimmers under warning, including at least swimmers in a first state who are presumed to be in a first drowning state where they are submerged, and swimmers in a second state who are presumed to be in a second drowning state, which is a different drowning state from the first drowning state and involves bodily movements, in a manner that distinguishes between the first drowning state and the second drowning state.

Claims

1. An image acquisition means for acquiring surveillance images of a swimming area taken from above the water surface, Information acquisition means for acquiring swimming state information output from a trained model by inputting the acquired monitoring image, A generation means for generating swimmer drowning warning information based on the acquired swimming status information, Equipped with, The swimming status information includes detection information of swimmers under suspicion, which includes at least swimmers in a first state, who are presumed to be in a first drowning state, submerged in water, and swimmers in a second state, who are in a drowning state different from the first drowning state and are presumed to be in a second drowning state, which involves bodily movements. The trained model is capable of distinguishing and detecting each of the multiple types of actions as the second drowning state in the monitoring image. The detection information that may be included in the swimming state information may include information that distinguishes between the first drowning state and the second drowning state, and that can identify each of the plurality of actions as the second drowning state. Swimmer monitoring device.

2. The aforementioned multiple types of actions that can be detected as the second drowning state include actions such as slapping the water surface, actions such as climbing a ladder, and actions in which the face or head comes in and out of the water surface. The swimmer monitoring device according to claim 1.

3. The trained model is capable of distinguishing and detecting each of several types of dangerous actions in the monitoring image that pose a risk of drowning. The swimmers under surveillance indicated by the detection information that may be included in the swimming status information further include swimmers who are presumed to be performing dangerous actions, The detection information that may be included in the swimming state information may include information that distinguishes between the first drowning state and the second drowning state and the dangerous action, and that can identify each of the multiple types of actions as the dangerous action. The swimmer monitoring device according to claim 1.

4. The aforementioned multiple types of actions that can be detected as dangerous actions include actions that involve moving while jumping up and down, actions that involve swimmers crossing paths with each other, and actions that involve diving around the depth adjustment platform. The swimmer monitoring device according to claim 3.

5. The swimming state information may include, in addition to the swimmers in the first state and the swimmers in the second state, detection information of swimmers under suspicion, further including swimmers in a third state who are presumed to be in a third drowning state, where they are floating with their faces submerged, in a manner that distinguishes between the first drowning state, the second drowning state, and the third drowning state. The swimmer monitoring device according to claim 1.

6. The detection information that may be included in the swimming status information further includes a probability value for each of the swimmers under surveillance, The generation means determines whether or not to generate the drowning warning information based on the probability value included in the detection information of the swimming state information. The swimmer monitoring device according to claim 1.

7. The trained model is capable of distinguishing and detecting swimmers under surveillance with floats and swimmers under surveillance without floats in the surveillance images. The detection information that may be included in the swimming status information further includes information that can distinguish between swimmers under surveillance with floats and swimmers under surveillance without floats. The swimmer monitoring device according to claim 1.

8. The trained model is capable of detecting mutually different behaviors in the monitoring image between the second drowning state of a second-state swimmer with a float and the second drowning state of a second-state swimmer without a float. The swimmer monitoring device according to claim 7.

9. The surveillance image acquired by the image acquisition means is a wide-angle image captured by a wide-angle camera. The trained model is trained using, as training data, at least the surveillance image of the swimmer in the first drowning state reflected in the central region, the surveillance image of the swimmer in the first partial region, and the surveillance image of the swimmer in the second partial region on the opposite side of the central region, and the surveillance image of the swimmer in the second drowning state reflected in the central region, the surveillance image of the swimmer in the first partial region, and the surveillance image of the swimmer in the second partial region. The swimmer monitoring device according to claim 1.

10. The surveillance image acquired by the image acquisition means is a wide-angle image captured by a wide-angle camera. The trained model is trained using, at least, the following as training data: the surveillance image of the first drowning swimmer reflected in the central region, the surveillance image of the first partial region, and the surveillance image of the second partial region on the opposite side of the first partial region across the central region; the surveillance image of the second drowning swimmer reflected in the central region, the surveillance image of the first partial region, and the surveillance image of the second partial region; and the surveillance image of the third drowning swimmer reflected in the central region, the surveillance image of the first partial region, and the surveillance image of the second partial region. The swimmer monitoring device according to claim 5.

11. One or more processors that can utilize the trained model, The process involves acquiring surveillance images of the swimming area taken from above the water surface, The process involves acquiring swimming state information output from the trained model by inputting the acquired monitoring image, A step of generating drowning warning information for swimmers based on the acquired swimming status information, Execute, The swimming status information includes detection information of swimmers under suspicion, which includes at least swimmers in a first state, who are presumed to be in a first drowning state, submerged in water, and swimmers in a second state, who are in a drowning state different from the first drowning state and are presumed to be in a second drowning state, which involves bodily movements. The trained model is capable of distinguishing and detecting each of the multiple types of actions as the second drowning state in the monitoring image. The detection information that may be included in the swimming state information may include information that distinguishes between the first drowning state and the second drowning state, and that can identify each of the plurality of actions as the second drowning state. Methods for monitoring swimmers.

Images