To solve the problem that in an image monitoring device for specifying an image of an object appearing in a monitored image, the state of occlusion generated between components due to the movement of the object changes and identification accuracy is reduced.

The object is represented by an object model 400 composed of a plurality of three-dimensional shapes (sub-models) for respective components. The object model 400 is virtually arranged in a monitoring space 70, the object model 400 is projected on an imaging surface 71 of an imaging section 2, and a sub-model visible region on which the sub-model is projected is determined for each sub-model. A part corresponding to the sub-model visible region is extracted as a component image of the object from the monitored image. Similarity between the monitored images is calculated based on the image feature quantity for each component. Integrated similarity integrated for the object is calculated, and is used for identifying the object.