We present a method for calculation of disparity maps from stereo sequences. Disparity map from previous frame is first transferred to the new frame using estimated motion of the calibrated stereo rig. The predicted disparities are validated for the new frame and areas where prediction failed are matched with a traditional stereo matching algorithm. This method produces very fast and temporally stable stereo matching suitable for real-time applications even on non-parallel hardware. IntroLast decade marked an increasing interest of researchers in stereo matching of image sequences. This task comes up mainly in automotive industry, 3D TV technologies and robotics.Using classical stereo algorithms (designed for standalone image pairs) on stereo sequences is not sufficient for these cases because of several issues. First of all, the resulting disparity maps are not temporally consistent -most methods exhibit unwanted flicker between the frames of the sequence. Second, it is desired to lower the computational complexity to achieve higher processing framerates. Typically this means that algorithms have higher error rate because they search a smaller volume in disparity space or do other simplifications. Finally, additional temporal information such as ego-motion or 3D scene flow may be extracted. Related Work Spacetime StereoEarly work on spacetime stereo [19,5] proposed extensions of spatial windows used for computation of matching costs to spatiotemporal windows, however they do not perform well with dynamic scenes. Their main advantage is that existing algorithms can be easily adapted to handle temporal dimension. Recent work from Richardt et al. [14] use spatiotemporal windows for the temporal variant of their algorithm (with addition of per-frame weights).Temporally stable stereo proposed in [15] considers image sequences as space-time volumes and the matching cost is based on similarity of spatiotemporal elements called stequels -optical flow is not explicitly computed to recover motion.A different approach is taken in [2] where temporal smoothing is applied as a post-processing of disparity maps using a median filter for each pixel over few adjacent frames. In order to cope with motion they compute optical flow between frames and trace the pixels over time.Min et al.[12] achieve temporal stability by adding a coherence function to the stereo matching cost to lower the matching cost in areas with small changes between frames. Stereo and Scene FlowThe concept of scene flow has been introduced in [17] as an extension of optical flow to temporal dimension. Some algorithms are designed to take advantage of joint calculation of disparity maps and disparity (scene) flow. For example,in [9] disparities are computed either using WTA or DP strategy, then the disparity flow is calculated using previous frame. Disparity prediction is done for the next frame and matching costs are updated to ensure temporal smoothness. In a similar fashion [3] describe a stereo algorithm with joint estimation of scene flow based o...

Data comprise intracranial EEG (iEEG) brain activity represented by stereo EEG (sEEG) signals, recorded from over 100 electrode channels implanted in any one patient across various brain regions. The iEEG signals were recorded in epilepsy patients (N = 10) undergoing invasive monitoring and localization of seizures when they were performing a battery of four memory tasks lasting approx. 1 hour in total. Gaze tracking on the task computer screen with estimating the pupil size was also recorded together with behavioral performance. Each dataset comes from one patient with anatomical localization of each electrode contact. Metadata contains labels for the recording channels with behavioral events marked from all tasks, including timing of correct and incorrect vocalization of the remembered stimuli. The iEEG and the pupillometric signals are saved in BIDS data structure to facilitate efficient data sharing and analysis.

In this contribution we examine the use and utility of parallel HMM classification in single-trial movement-EEG classification of index finger reaching and grasping movement. Parallel HMMs allow us to easily utilize the information contained in multiple channels. Using HMM classifier output in parallel from examined EEG channels we have been able to achieve as good a classification score as with single electrode results, further we do not rely on a single electrode giving persistently good results. Our parallel approach has the added benefit of not having to rely on small inter-session variability as it gives very good results with fewer classifier parameters being optimized. Without any classification optimization we can get a score improvement of 11.2% against randomly selected physiologically relevant electrode. If we use subject specific information we can further improve on the reference score by 1%, achieving a classification score of 84.2±0.7%.

The contribution investigates the impact of frequency feature optimization on discriminating between movement-related EEG realisations associated with right shoulder elevation and right index finger flexion movements. Exhaustive search of subbands in the range from 5 to 45 Hz is performed. A classifier based on Hidden Markov Models is utilised. The results show a large variability of optimal settings among subjects and electrodes. Using subband optimization an average 3.5% increase in classification accuracy of EEG filtered using 8-neighbor Laplacian filter was achieved, reaching an overall score of 81.2±1.2%, individual improvements ranging from 1.2 to 9.9%. The best general setting common for all subject was confirmed as 5-40 Hz.