Multiresolution ORKA: fast and resolution independent object reconstruction using a K-approximation graph

Object recognition and reconstruction is of great interest in many research fields. Detecting pedestrians or cars in traffic cameras or tracking seismic waves in geophysical exploration are only two of many applications. Recently, the authors developed a new method - Object reconstruction using K-approximation (ORKA) - to extract such objects out of given data. In this method a special object model is used where the movement and deformation of the object can be controlled to fit the application. ORKA in its current form is highly dependent on the data resolution. On the one hand, the movement of the object can only be reconstructed on a grid that depends on the data resolution. On the other hand, the runtime increases exponentially with the resolution. Hence, the resolution of the data needs to be in a small range where the reconstruction is accurate enough but the runtime is not too high. In this work, we present a multiresolution approach, where we combine ORKA with a wavelet decomposition of the data. The object is then reconstructed iteratively what drastically reduces the runtime. Moreover, we can increase the data resolution such that the movement reconstruction no longer depends on the original grid. We also give a brief introduction on the original ORKA algorithm. Hence, knowledge of the previous work is not required.