An image processing device includes a detector, a setting unit, a calculator, an estimator, and a determining unit. The detector sets a three-dimensional object region by detecting a three-dimensional object. The setting unit sets a first image region corresponding to the three-dimensional object region, a second image region including a left end of and partly overlapping the first image region, and a third image region including a right end of and partly overlapping the first image region. The calculator calculates representative values of parallax-related values in pixel columns in the image regions, and calculates an approximate line of the representative values in each of the image regions. The estimator estimates a continuous structure degree, from a slope value of the approximate line of each of the image regions. The determining unit determines whether the three-dimensional object is a vehicle, from the continuous structure degree.