Abstract-The accuracy evaluation of image feature detectors is done using the repeatability criterion. Therefore, a wellknown data set consisting of image sequences and homography matrices is processed. This data serves as ground truth mapping information for the evaluation and is used in many computer vision papers.An accuracy validation of the benchmarks has not been done so far and is provided in this work. The accuracy is limited and evaluations of feature detectors may result in erroneous conclusions.Using a differential evolution approach for the optimization of a new, feature-independent cost function, the accuracy of the ground truth homographies is increased. The results are validated using comparisons between the repeatability rates before and after the proposed optimization. The new homographies provide better repeatability results for each detector. The repeatability rate is increased by up to 20%.

In recent decades, cold atom experiments have become increasingly complex. While computers control most parameters, optimization is mostly done manually. This is a time-consuming task for a high-dimensional parameter space with unknown correlations. Here we automate this process using a genetic algorithm based on Differential Evolution. We demonstrate that this algorithm optimizes 21 correlated parameters and that it is robust against local maxima and experimental noise. The algorithm is flexible and easy to implement. Thus, the presented scheme can be applied to a wide range of experimental optimization tasks.Comment: minor revisio

Abstract. Benchmark data sets consisting of image pairs and ground truth homographies are used for evaluating fundamental computer vision challenges, such as the detection of image features. The mostly used benchmark provides data with only low resolution images. This paper presents an evaluation benchmark consisting of high resolution images of up to 8 megapixels and highly accurate homographies. State of the art feature detection approaches are evaluated using the new benchmark data. It is shown that existing approaches perform differently on the high resolution data compared to the same images with lower resolution.

Abstract. Common techniques in structure from motion do not explicitly handle foreground occlusions and disocclusions, leading to several trajectories of a single 3D point. Hence, different discontinued trajectories induce a set of (more inaccurate) 3D points instead of a single 3D point, so that it is highly desirable to enforce long continuous trajectories which automatically bridge occlusions after a re-identification step. The solution proposed in this paper is to connect features in the current image to trajectories which discontinued earlier during the tracking. This is done using a correspondence analysis which is designed for wide baselines and an outlier elimination strategy using the epipolar geometry. The reference to the 3D object points can be used as a new constraint in the bundle adjustment. The feature localization is done using the SIFT detector extended by a Gaussian approximation of the gradient image signal. This technique provides the robustness of SIFT coupled with increased localization accuracy.Our results show that the reconstruction can be drastically improved and the drift is reduced, especially in sequences with occlusions resulting from foreground objects. In scenarios with large occlusions, the new approach leads to reliable and accurate results while a standard reference method fails.