Physical adversarial attacks against vehicle detection are gaining attention. However, most prior works focus on improving attack ability by amplifying the intensity and scope of perturbations, which results in a visually suspicious appearance that exposes attackers’ behavior. Motivated by the shape preference characteristic exhibited in human cognitive processes, we propose a shape constraint physical camouflage attack (SC-PCA) to generate vehicle camouflage. To generate naturalistic perturbations, we use a contour image as the control condition and introduce a shape-aware loss in conditional generative adversarial network. Then, we map the perturbations onto the surface of the target vehicle to form the camouflage. By setting different transformation parameters, vehicle images of multiple perspectives and multiple scenes can be rendered. Experiments conducted in both digital and physical worlds demonstrate that our method has a good attack ability, can deceive the vehicle detector in the real world, and can adapt to various angle, distance, and background changes. Moreover, the outcome of the human perception survey indicates that our approach outperforms the state-of-the-art techniques.