An image analysis apparatus picks up an image of a region ahead of a vehicle using a camera, and allows a control unit to analyze picked-up image data generated by the camera to learn a focus-of-expansion position. The control unit controls the learning performance as follows. Specifically, the control unit does not start the learning performance for the focus-of-expansion position until a state where a detection value of a vehicle speed exceeds a reference speed exceeds a specified duration of time. When the state where a detection value of a vehicle speed exceeds a reference speed exceeds the specified duration of time, the learning performance for the focus-of-expansion position is started from this time point. The specified duration of time may be determined on the basis of statistics on the durations of simulated runs of the vehicle performed on a chassis dynamometer in a vehicle inspection.