Publikationen

Surround View fisheye cameras are commonly deployed in automated driving for 360˚ near-field sensing around the vehicle. This work presents a multi-task visual perception network on unrectified fisheye images to enable the vehicle to sense its surrounding environment. It consists of six primary tasks necessary for an autonomous driving system: depth estimation, visual odometry, semantic segmentation, motion segmentation, object detection, and lens soiling detection. We demonstrate that the jointly trained model performs better than the respective single task versions. Our multi-task model has a shared encoder providing a significant computational advantage and has synergized decoders where tasks support each other. We propose a novel camera geometry based adaptation mechanism to encode the fisheye distortion model both at training and inference. This was crucial to enable training on the WoodScape dataset, comprised of data from different parts of the world collected by 12 different cameras mounted on three different cars with different intrinsics and viewpoints. Given that bounding boxes is not a good representation for distorted fisheye images, we also extend object detection to use a polygon with non-uniformly sampled vertices. We additionally evaluate our model on standard automotive datasets, namely KITTI and Cityscapes. We obtain the state-of-the-art results on KITTI for depth estimation and pose estimation tasks and competitive performance on the other tasks. We perform extensive ablation studies on various architecture choices and task weighting methodologies. A short video at https://youtu.be/xbSjZ5OfPes provides qualitative results.

This paper deals with the controller design of a linear drive system with potential applications in picometer-scale positioning. The displacement is measured using a high-precision differential plane-mirror laser interferometer. However, this measured signal must be filtered for control purposes and performance validation. For tracing picometers, we adopt a model-based 2-DOF control architecture with a feedforward stage for model-following and a feedback stage for stabilization and compensation of constant disturbances. A straightforward way to deal with noise is to use a feedback controller based on estimated (and filtered) states, which can be generated by a Kalman-Bucy filter. The effectiveness of the presented control strategy is verified via real-time experimentation, where the overall control scheme allows positioning in the picometer level. Moreover, comparative tests in the frequency domain were conducted using the so-called PI-LQR feedback controller based on estimated states.

Purpose The localized heating of magnetic nanoparticles (MNPs) via the application of time-varying magnetic fields - a process known as magnetic field hyperthermia (MFH) - can greatly enhance existing options for cancer treatment; but for broad clinical uptake its optimization, reproducibility and safety must be comprehensively proven. As part of this effort, the quantification of MNP heating - characterized by the specific loss power (SLP), measured in W/g, or by the intrinsic loss power (ILP), in Hm2/kg - is frequently reported. However, in SLP/ILP measurements to date, the apparatus, the analysis techniques and the field conditions used by different researchers have varied greatly, leading to questions as to the reproducibility of the measurements.Materials and Methods An interlaboratory study (across N = 21 European sites) of calorimetry measurements that constitutes a snapshot of the current state-of-the-art within the MFH community has been undertaken. Identical samples of two stable nanoparticle systems were distributed to all participating laboratories. Raw measurement data as well as the results of in-house analysis techniques were collected along with details of the measurement apparatus used. Raw measurement data was further reanalyzed by universal application of the corrected-slope method to examine relative influences of apparatus and results processing.Results The data show that although there is very good intralaboratory repeatability, the overall interlaboratory measurement accuracy is poor, with the consolidated ILP data having standard deviations on the mean of ca. ± 30% to ± 40%. There is a strong systematic component to the uncertainties, and a clear rank correlation between the measuring laboratory and the ILP. Both of these are indications of a current lack of normalization in this field. A number of possible sources of systematic uncertainties are identified, and means determined to alleviate or minimize them. However, no single dominant factor was identified, and significant work remains to ascertain and remove the remaining uncertainty sources.Conclusion We conclude that the study reveals a current lack of harmonization in MFH characterization of MNPs, and highlights the growing need for standardized, quantitative characterization techniques for this emerging medical technology.