Feature matching is at the base of the target detection problem. Current methods rely on costly descriptors for detection and matching. This paper presents an improved feature descriptor based on ORB, called Gravity-ORB for target detection in mobile devices. Compared with traditional descriptors such as SIFT or ORB, the concept design can perform fast feature matching under the condition of keeping the restriction on robustness, even in the case where mobile devices have limited computational capacity. Specially, Gravity-ORB reduces the complexity of feature computation for mobile devices with less computing by using gravity acceleration sensors. In the end, experiments conducted in smart phones and tablets demonstrate the effectiveness and real-time performance of the proposed method.