The subject matter of this specification can be implemented in, among other things, a computer-implemented method for object detection. The method includes detecting at least one initial position of an object of a specific type in at least one initial image within an ordered sequence of images stored in a memory. The method further includes detecting at least one subsequent position of the object in at least one subsequent image within the ordered sequence of images. The method further includes estimating one or more intermediate positions of the object in one or more intermediate images within the ordered sequence of images between the initial image and the subsequent image based on the initial position and the subsequent position.