Ubiquitous mobile devices are used for various tasks in computer vision. Wagner et al. perform robust 6DOF natural feature tracking using modified SIFT and Ferns as descriptors [11]. In order to allow computation on mobile devices, extensive modifications are carried out to the basic concepts of SIFTand Ferns,followed by a instructive evaluation of system performance. Despite severe limitations in processing speed and memory bandwidth, they achieve real-time performance when using textured planar targets on current-generation phones. Klein and Murray present a system for parallel tracking and mapping on camera phones [7]. They implement a key-frame based SLAM system that is capable of generating and augmenting small maps. Limited computational resources and problems in image acquisition such as a rolling shutter a respecifically accounted for to allow computation on mobile devices (iPhone 3G). In the system at hand robust estimation of object features from a single input image must be performed. Due to the nature of the problem, it seems justified to run through a separated segmentation, feature extraction and classification step (depending on the feature), in which tasks may be optimised independently. For each of the steps needed in our application, a vast amount of literature is available, but self-contained work on how to perform these steps efficiently on mobile devices is not available. Thus, we propose a series of solutions which make the problem computable on current mobile phones in instant time.