According to an embodiment, an image processing device includes an image acquirer, a tracker, an extractor, a feature amount acquirer, and an association controller. The image acquirer is configured to acquire a time series of images from each imaging device. The tracker is configured to track a position of an object in the time series of images. The extractor is configured to extract, from the time series of images, an observed value corresponding to a feature value in each of regions into which the time series of images is divided depending on a distance from the imaging device. The feature amount acquirer is configured to acquire a feature amount based on a deviation value of the observed value in a feature value distribution for the region. The association controller is configured to associate the tracked objects among multiple time series of images acquired from the imaging devices by using the feature amount.