Endpoint Error Research Articles

An Edge-Sense Bidirectional Pyramid Network for Stereo Matching of VHR Remote Sensing Images

As an essential step in 3D reconstruction, stereo matching still faces unignorable problems due to the high resolution and complex structures of remote sensing images. Especially in occluded areas of tall buildings and textureless areas of waters and woods, precise disparity estimation has become a difficult but important task. In this paper, we develop a novel edge-sense bidirectional pyramid stereo matching network to solve the aforementioned problems. The cost volume is constructed from negative to positive disparities since the disparity range in remote sensing images varies greatly and traditional deep learning networks only work well for positive disparities. Then, the occlusion-aware maps based on the forward-backward consistency assumption are applied to reduce the influence of the occluded area. Moreover, we design an edge-sense smoothness loss to improve the performance of textureless areas while maintaining the main structure. The proposed network is compared with two baselines. The experimental results show that our proposed method outperforms two methods, DenseMapNet and PSMNet, in terms of averaged endpoint error (EPE) and the fraction of erroneous pixels (D1), and the improvements in occluded and textureless areas are significant.

An Acute Application of Cerebellar Transcranial Direct Current Stimulation Does Not Improve Motor Performance in Parkinson's Disease.

Transcranial direct current stimulation of the cerebellum (c-tDCS) improves motor performance in young and old adults. Based on the cerebellar involvement in Parkinson’s disease (PD), c-tDCS could have potential to improve motor function in PD. The purpose was to determine the effects of c-tDCS on motor performance in PD while participants were on medications. The study was a randomized, double-blind, SHAM-controlled, between-subjects design. Twenty-two participants with PD were allocated to either a c-tDCS group or a SHAM group. All participants completed one experimental session and performed two motor tasks with their most affected hand in a Baseline condition (no stimulation) and an Experimental condition. The motor tasks were a visuomotor isometric precision grip task (PGT) and a rapid arm movement task (AMT). The primary dependent variables were force error and endpoint error in the PGT and AMT, respectively. There were no significant differences in force error or endpoint error in the Experimental condition between the c-tDCS and SHAM groups. These results indicate that an acute application of c-tDCS does not enhance motor performance in hand and arm tasks in PD. Longer-term c-tDCS application over multiple days may be needed to enhance motor function in PD.

A Direct Drive Parallel Plane Piezoelectric Needle Positioning Robot for MRI Guided Intraspinal Injection.

To improve both the procedure time and positioning accuracy, an MRI guided robotic needle positioning system is developed. The robot uses linear piezoelectric motors to directly drive a parallel plane positioning mechanism. Feedback is provided through MRI during the orientation procedure. Both accuracy and repeatability of the robot are characterized. This system is found to be capable of repeatability below 51μm. Needle endpoint error is limited by imaging modality, but is validated to 156μm. The reported robot and MRI image feedback system is capable of repeatable and accurate needle guide positioning. This high accuracy will result in a significant improvement to the workflow of spinal injection procedures.

Convolutional neural network-based dosimetry evaluation of esophageal radiation treatment planning.

A dosimetry evaluation model for treatment planning of esophageal radiation therapy is developed using a deep learning model. The model predicts dose volume histogram (DVH) from distance to target histogram (DTH) based on stacked de-noise auto-encoder (SDAE) and one-dimensional convolutional network (1D-CN). First, SDAE is used to extract the features from the curves of DTH and DVH. Then 1D-CN model is employed to learn the relationship between the features of DTH and DVH, and later used to predict the features of DVH from the features of DTH. Finally, the curve of DVH is restored from the features of DVH based on SDAE. Two hundred and seventy treatment plans are used for training 1D-CN and another sixty-three treatment plans are used for evaluating this model. This method is also compared with another two popular prediction methods based on support vector machine (SVM) and U-net. Based on the experimental result, the proposed model achieves the lowest dose endpoint error comparing to the other models. The average prediction error on planned target volume, left lung, right lung, heart, and spinal cord is 2.94% for the proposed model, while the average prediction errors are 6.79% and 3.41% for SVM and U-net, respectively. A dosimetry evaluation method based on SDAE and 1D-CN is developed in characterizing the correlation relationship between DTH and DVH of treatment plans. The results show that the model could be trained more efficiently in this framework and the DVH could be predicted with higher accuracy comparing to those existing methods. It provides a useful tool in supporting automated treatment planning of esophageal intensity-modulated radiotherapy.

Lower extremity prism adaptation in individuals with anterior cruciate ligament reconstruction

BackgroundEmerging research has proposed a growing reliance on visual processing during motor performance in individuals following anterior cruciate ligament reconstruction. Reconstructed individuals display increased activation of visual processing areas during task execution and exhibit dramatic performance decrements when vision is completely removed, however the effect of visual information manipulation on performance remains unknown. The purpose of this study was to determine how manipulation of visual information changes performance in persons with anterior cruciate ligament reconstruction. MethodsTwenty-one persons with anterior cruciate ligament reconstruction and 21 matched healthy adults reached to a target with the toe of the involved limb 50 times while wearing prism goggles that vertically shifted their visual field. Toe kinematics were collected to quantify endpoint error and reaching behavior. FindingsStatistical analyses failed to detect significant differences, evidencing both groups performed similarly with respect to endpoint error, movement duration, peak and maximum endpoint velocities, and initial direction error. InterpretationWhen provided inaccurate information via a visual field perturbation, both groups demonstrated comparable adaptation and post-adaptation behavior. These results suggest this sample of persons with anterior cruciate ligament reconstruction are able to effectively integrate information across sensory systems as well as non-injured individuals.

DENSE 3D OBJECT RECONSTRUCTION USING STRUCTURED-LIGHT SCANNER AND DEEP LEARNING

Abstract. Structured light scanners are intensively exploited in various applications such as non-destructive quality control at an assembly line, optical metrology, and cultural heritage documentation. While more than 20 companies develop commercially available structured light scanners, structured light technology accuracy has limitations for fast systems. Model surface discrepancies often present if the texture of the object has severe changes in brightness or reflective properties of its texture. The primary source of such discrepancies is errors in the stereo matching caused by complex surface texture. These errors result in ridge-like structures on the surface of the reconstructed 3D model. This paper is focused on the development of a deep neural network LineMatchGAN for error reduction in 3D models produced by a structured light scanner. We use the pix2pix model as a starting point for our research. The aim of our LineMatchGAN is a refinement of the rough optical flow A and generation of an error-free optical flow B̂. We collected a dataset (which we term ZebraScan) consisting of 500 samples to train our LineMatchGAN model. Each sample includes image sequences (Sl, Sr), ground-truth optical flow B and a ground-truth 3D model. We evaluate our LineMatchGAN on a test split of our ZebraScan dataset that includes 50 samples. The evaluation proves that our LineMatchGAN improves the stereo matching accuracy (optical flow end point error, EPE) from 0.05 pixels to 0.01 pixels.

METHOD OF OPTICAL FLOW CALCULATION BASED ON FEATURE AUGMENTATION

The paper addresses the problem of robust optical flow computation for real-time applications. We present an efficient, in presence of varying illumination and large displacements, method of dense optical flow calculation based on Robust Local Optical Flow (RLOF). Introduction of Adaptive and Generic Accelerated Segment Test (AGAST) feature detector with additional augmentation by simple grid points provides a stable and reliable under difficult conditions set of features and uniform points that produces dense and accurate optical flow and leads to better performance. The usage of feature points for prior initialization leads for robustness for large displacements as well. The other essential step of optical flow calculation – densification of optical flow – was strengthen by modification of weights. The raster interpolation was applied with weights on second derivatives to make more edgesensitive optical flow. To enhance the overall result, a post-processing step of variational refinement process were added to optical flow calculation framework. The proposed method was evaluated on Max Planck Institute (MPI) Sintel dataset – a challenging set of 3D-animated movies with large range of motions and illuminations, which has ground truth optical flow available. The comparison with initial RLOF method were measured by running time and quality parameters, measured by End Point Error (EPE) and Angular Error (AE). The evaluation shows the decrease of running time by approximately 40 % and improvement of EPE and AE, that demonstrates the effectiveness of the method.

Performance of discrete, reciprocal, and cyclic movements of the ipsilesional upper limb in individuals after stroke.

Aiming movements of the upper limbs can be classified either as discrete, or reciprocal, or cyclic. The control of these movements after a stroke can be affected. The aim of this experimental, cross-sectional study was to characterize the performance of these movements after the right and left hemisphere chronic stroke. Thirty-six individuals aged between 40 and 70years, right-handed, were allocated into three groups (control, right stroke, and left stroke). Participants were asked to perform aiming movements on a tablet. Individuals after stroke performed the tasks only with their ipsilesional limb, while the control group performed movements with both limbs. The reaction and movement times, peak velocity, and the variability and error of the endpoint were analyzed. Individuals after stroke presented a worse performance in all movement classes as expected, but differently depending on the damaged hemisphere. Participants with right hemisphere damage showed larger endpoint errors, while those with left hemisphere damage had longer reaction and movement times. Both differences were seen consistently in discrete and reciprocal, but not in cyclic movements. Cyclic movements presented shorter latencies, were faster, and showed greater endpoint errors when compared to discrete and reciprocal movements. These results suggest that stroke affects differently the performance of discrete and reciprocal movements according to the hemisphere lesion side, but not in cyclic movements. Different levels of motor control among the three classes of movements by the nervous system may justify these results.

BlockNet: A Deep Neural Network for Block-Based Motion Estimation Using Representative Matching

Owing to the limitations of practical realizations, block-based motion is widely used as an alternative for pixel-based motion in video applications such as global motion estimation and frame rate up-conversion. We hereby present BlockNet, a compact but effective deep neural architecture for block-based motion estimation. First, BlockNet extracts rich features for a pair of input images. Then, it estimates coarse-to-fine block motion using a pyramidal structure. In each level, block-based motion is estimated using the proposed representative matching with a simple average operator. The experimental results show that BlockNet achieved a similar average end-point error with and without representative matching, whereas the proposed matching incurred 18% lower computational cost than full matching.

Rhythmic auditory stimuli heard before and during a reaching movement elicit performance improvements in both temporal and spatial movement parameters

Rhythmic auditory stimuli (RAS) have been proposed to improve motor performance in populations with and without sensorimotor impairments. However, the reasons for the reported benefits are poorly understood. One idea is that RAS may supplement intrinsic feedback when other sensory input is diminished. The current experiment tested this idea by removing vision during a goal-directed reaching task. We hypothesized that any improvements in movement performance due to the RAS would be greater when vision was removed. Twenty-two typically developing adults performed reaching movements to one of two targets with RAS presented before movement initiation, after movement initiation, both before and after movement initiation, and no sound, all with and without vision. Dependent variables were analyzed using a 2 vision by 2 sound-before by 2 sound-during repeated measures ANOVA. Conditions where the metronome was heard before movement initiation yielded shorter and less variable reaction times compared when there was no sound before the movement. The RAS heard before and during the movement independently impacted spatial aspects of the movement. Sound before movement initiation resulted in smaller endpoint error, primarily in the anterior-posterior axis. Sound during the movement resulted in smaller endpoint error, primarily in the mediolateral axis. In no-vision blocks, inclusion of RAS resulted in improved endpoint performance, indicating that RAS supplemented the motor system. The present results strengthen our understanding of sensory integration underlying reaching performance by demonstrating that sound heard before and during a reaching movement can improve motor performance by supplementing the motor system when vision is unavailable.

Model identification of a Marine robot in presence of IMU-DVL misalignment using TUKF

In today’s world, control and navigation of autonomous underwater vehicles (AUVs) are quite challenging issues. In these fields, obtaining an identified dynamic model of AUV is a vital part. In this paper, a method for parameter estimation of an AUV planar model is proposed, which uses augmented state space technique and Square Root Transformed Unscented Kalman Filter (SR-TUKF) as an estimator. Furthermore, by modeling, misalignment between Inertial Measurement Unit (IMU) and Doppler Velocity Log (DVL) is estimated, simultaneously. Parameter identification is conducted using data of an AUV, equipped with Gyroscope, DVL and Encoder for measuring control inputs, in a planar maneuver. According to the experimental results, the output of the identified model has matched appropriately to the output of sensors in validation maneuvers. With and without considering the estimated misalignment between IMU-DVL, navigation is performed by fusing inertial navigation system (INS) with DVL. The results show an improvement in the end point error with respect to the travel distance by 53 percent in a maneuver with a distance of 6557 meters. Therefore, the proposed method for estimation of a planar model of AUV and misalignment of sensors has a plausible result, experimentally verified by raw data of a simple maneuver.

Configuration Optimization for End-Point Stabilization of Redundant Manipulators With Base Flexibility

Abstract This paper presents a method to determine an optimal configuration of a teleoperated excavator to minimize the induced undercarriage oscillation for robust end-point stabilization. Treating the excavator as a kinematically redundant system, where non-unique combinations of the undercarriage position and arm posture can locate the end-point at the same reference. A specific configuration can be chosen to not excite undercarriage oscillation with simple end-point error feedback control without model-based or measurement-based vibration suppression. Robust stability measures based on normalized coprime factorization as well as modal decomposition solve the redundancy of the kinematics. An advantage of this approach is that the control engineer can proceed as if the excavator arm is fixed to rigid ground, which is practically not the case, and apply simple traditional Jacobian-based end-point control.

High accuracy correspondence field estimation via MST based patch matching

This paper presents an effective framework for correspondence field estimation. The core idea is to construct pixel-level and superpixel-level patch matching to achieve high accuracy estimation as well as fast speed computation. To this end, a hybrid edge-preserving supported weighting approach is first developed, which contributes to better performance on the pixel level, especially on those in the regions of fine structures. Then, a local Minimum Spanning Tree (MST) is constructed to describe regions and develop the adaptive smooth penalty weights, so that the over-patching in large textureless regions can be effectively avoided. In addition, the MST is further extended to handle occlusions in way of edge preserving strategy. Finally, all the above treatments are collected into an optimization model where the objective function is developed in terms of Markov Random Filed (MRF). In computation, a fast yet efficient iterative optimization strategy is developed. Our approach achieves favorable place on optical flow benchmark, which locates at the top two and top four for endpoint error and angular error evaluations among more than 130 approaches listed in the webpage.

Temporal Invariance in SCA6 Is Related to Smaller Cerebellar Lobule VI and Greater Disease Severity.

Regulating muscle force and timing are fundamental for accurate motor performance. In spinocerebellar ataxia type 6 (SCA6), there is evidence that individuals have greater force dysmetria but display better temporal accuracy during fast goal directed contractions. Here, we test whether greater temporal accuracy occurs in all individuals with SCA6, and can be explained by lesser temporal variability. Further we examine whether it is linked to disease severity and specific degenerative changes in the cerebellum. Nineteen human participants with SCA6 (13 woman) and 18 healthy controls performed fast goal-directed ankle dorsiflexion contractions aiming at a spatiotemporal target. We quantified the endpoint control of these contractions, gray matter (GM) integrity of the cerebellum, and disease severity using the International Cooperative Ataxia Rating Scale (ICARS). SCA6 individuals exhibited lower temporal endpoint error and variability than the healthy controls (p = 0.008). Statistically, SCA6 clustered into two distinct groups for temporal variability. A group with low temporal variability ranging from 10 to 19% (SCA6a) and a group with temporal variability similar to healthy controls (SCA6b; 19-40%).SCA6a exhibited greater disease severity than SCA6b, as assessed with ICARS (p < 0.001). Lower temporal variability, which was not associated with disease duration (R2 = 0.1, p > 0.2), did correlate with both greater ICARS (R2 = 0.3) and reduced GM volume in cerebellar lobule VI (R2 = 0.35). Other cerebellar lobules did not relate to temporal variability. We provide new evidence that a subset of SCA6 with greater loss of GM in cerebellum lobule VI exhibit temporal invariance and more severe ataxia than other SCA6 individuals.SIGNIFICANCE STATEMENT Variability is an inherent feature of voluntary movement, and traditionally more variability in the targeted output infers impaired performance. For example, cerebellar patients present exacerbated temporal variability during multijoint movements, which is thought to contribute to their motor deficits. In the current work, we show that in a subgroup of spinocerebellar ataxia type 6 individuals, temporal variability is lower than that of healthy controls when performing single-joint fast-goal directed movements. This invariance related to exacerbated atrophy of lobule VI of the cerebellum and exacerbated disease severity. The relation between invariance and disease severity suggests that pathological motor variability can manifest not only as an exacerbation but also as a reduction relative to healthy controls.

Exponentially-fitted pseudo Runge-Kutta method

This article is devoted to the development of an embedded pseudo-Runge-Kutta method of order three (EPRK3) and its exponential-fitting. The motivation behind the development is to minimise the cost of computation for existing Runge-Kutta (RK) type method and also to make EPRK method compatible to solve the initial value problem (IVP) having periodic solutions. Here we assume that ef-EPRKM exactly integrates two exponential functions e±ωx, with unknown frequency ω. The proposed methods are applied to two IVPs of order two. The computation cost in terms of total function evaluations is compared in Table 1. In Tables 2 and 3, a comparison of norms of endpoint errors is made between EPRK3 method, ef-EPRK3 method, Berghe's ef-RK method and Simos's ef-RK method from which it is quite evident that errors by ef-EPRKM are smallest. To compute unknown frequency ω in e±ωx, the local truncation error (LTE) for ef-EPRKM is computed.

End-to-End Speech Endpoint Detection Utilizing Acoustic and Language Modeling Knowledge for Online Low-Latency Speech Recognition

Speech endpoint detection (EPD) benefits from the decoder state features (DSFs) of online automatic speech recognition (ASR) system. However, the DSFs are obtained via the ASR decoding process, which can become prohibitively expensive especially in limited-resource scenarios such as the embedded devices. To address this problem, this paper proposes a language model (LM)-based end-of-utterance (EOU) predictor, which is trained to determine the framewise probabilities of the EOU token conditioned on the previous word history obtained from the 1-best decoding hypothesis of the ASR system in an end-to-end manner without an actual decoding process in the test step. Further, a novel end-to-end EPD strategy is presented to incorporate a phonetic embedding (PE)-based acoustic modeling knowledge and the proposed EOU predictor-based language modeling knowledge into an acoustic feature embedding (AFE)-based EPD approach within the recurrent neural networks (RNN)-based EPD framework. The proposed EPD algorithm is built upon the ensemble RNNs, which are independently trained for the three parts, which are the proposed LM-based EOU predictor, AFE-based EPD, and PE-based acoustic model (AM) in accordance with each target. The ensemble RNNs are concatenated at the level of the last hidden layers and then attached into the fully-connected deep neural networks (DNN)-based EPD classifier, which is trained in accordance with the ultimate EPD target. Thereafter, they are jointly retrained at the second step of the DNN training to yield the lower endpoint error. The proposed EPD framework was evaluated in terms of the endpoint accuracy and word error rate for the CHiME-3 and large-scale ASR tasks. The experimental results turn out that the proposed EPD algorithm efficiently outperforms the conventional EPD approaches.

Exponentially-fitted pseudo Runge-Kutta method

This article is devoted to the development of an embedded pseudo-Runge-Kutta method of order three (EPRK3) and its exponential-fitting. The motivation behind the development is to minimise the cost of computation for existing Runge-Kutta (RK) type method and also to make EPRK method compatible to solve the initial value problem (IVP) having periodic solutions. Here we assume that ef-EPRKM exactly integrates two exponential functions e±ωx, with unknown frequency ω. The proposed methods are applied to two IVPs of order two. The computation cost in terms of total function evaluations is compared in Table 1. In Tables 2 and 3, a comparison of norms of endpoint errors is made between EPRK3 method, ef-EPRK3 method, Berghe's ef-RK method and Simos's ef-RK method from which it is quite evident that errors by ef-EPRKM are smallest. To compute unknown frequency ω in e±ωx, the local truncation error (LTE) for ef-EPRKM is computed.

Explanation of Fitts\u2019 law in Reaching Movement based on Human Arm Dynamics

Why does Fitts’ law fit various human behavioural data well even though it is not a model based on human physical dynamics? To clarify this, we derived the relationships among the factors applied in Fitts’ law—movement duration and spatial endpoint error—based on a multi-joint forward- and inverse-dynamics models in the presence of signal-dependent noise. As a result, the relationship between them was modelled as an inverse proportion. To validate whether the endpoint error calculated by the model can represent the endpoint error of actual movements, we conducted a behavioural experiment in which centre-out reaching movements were performed under temporal constraints in four directions using the shoulder and elbow joints. The result showed that the distributions of model endpoint error closely expressed the observed endpoint error distributions. Furthermore, the model was found to be nearly consistent with Fitts’ law. Further analysis revealed that the coefficients of Fitts’ law could be expressed by arm dynamics and signal-dependent noise parameters. Consequently, our answer to the question above is: Fitts’ law for reaching movements can be expressed based on human arm dynamics; thus, Fitts’ law closely fits human’s behavioural data under various conditions.

Robust optical flow estimation involving exponential fractional-order derivatives

This work presents a fractional-order method for optical flow estimation. The proposed method is based on the Classic+NL algorithm and the Caputo-Fabrizio fractional-order derivative. Therefore, the proposed operator can be seen as a generalization of this integer-order model. The performance of the algorithm is analyzed by using the Average End Point Error (EPE) and the Average Angular Error (AAE) considering the MPI Sintel and Middlebury Datasets. Experimental results on these dataset show that the proposed algorithm performs better than other existing techniques through the analysis of the two error metrics mentioned above. The firefly optimization algorithm was applied to obtain the fractional-order, which minimizes the EPE, for each of images sequence used.

No effect of triple-pulse TMS medial to intraparietal sulcus on online correction for target perturbations during goal-directed hand and foot reaches.

Posterior parietal cortex (PPC) is central to sensorimotor processing for goal-directed hand and foot movements. Yet, the specific role of PPC subregions in these functions is not clear. Previous human neuroimaging and transcranial magnetic stimulation (TMS) work has suggested that PPC lateral to the intraparietal sulcus (IPS) is involved in directing the arm, shaping the hand, and correcting both finger-shaping and hand trajectory during movement. The lateral localization of these functions agrees with the comparably lateral position of the hand and fingers within the motor and somatosensory homunculi along the central sulcus; this might suggest that, in analogy, (goal-directed) foot movements would be mediated by medial portions of PPC. However, foot movement planning activates similar regions for both hand and foot movement along the caudal-to-rostral axis of PPC, with some effector-specificity evident only rostrally, near the central regions of sensorimotor cortex. Here, we attempted to test the causal involvement of PPC regions medial to IPS in hand and foot reaching as well as online correction evoked by target displacement. Participants made hand and foot reaches towards identical visual targets. Sometimes, the target changed position 100–117 ms into the movement. We disturbed cortical processing over four positions medial to IPS with three pulses of TMS separated by 40 ms, both during trials with and without target displacement. We timed TMS to disrupt reach execution and online correction. TMS did not affect endpoint error, endpoint variability, or reach trajectories for hand or foot. While these negative results await replication with different TMS timing and parameters, we conclude that regions medial to IPS are involved in planning, rather than execution and online control, of goal-directed limb movements.