A Framework for Representing, Building and Reusing Novel State-of-the-Art Three-Dimensional Object Detection Models in Point Clouds Targeting Self-Driving Applications.

The rapid development of deep learning has brought novel methodologies for 3D object detection using LiDAR sensing technology. These improvements in precision and inference speed performances lead to notable high performance and real-time inference, which is especially important for self-driving purposes. However, the developments carried by these approaches overwhelm the research process in this area since new methods, technologies and software versions lead to different project necessities, specifications and requirements. Moreover, the improvements brought by the new methods may be due to improvements in newer versions of deep learning frameworks and not just the novelty and innovation of the model architecture. Thus, it has become crucial to create a framework with the same software versions, specifications and requirements that accommodate all these methodologies and allow for the easy introduction of new methods and models. A framework is proposed that abstracts the implementation, reusing and building of novel methods and models. The main idea is to facilitate the representation of state-of-the-art (SoA) approaches and simultaneously encourage the implementation of new approaches by reusing, improving and innovating modules in the proposed framework, which has the same software specifications to allow for a fair comparison. This makes it possible to determine if the key innovation approach outperforms the current SoA by comparing models in a framework with the same software specifications and requirements.

Customizable FPGA-Based Hardware Accelerator for Standard Convolution Processes Empowered with Quantization Applied to LiDAR Data.

In recent years there has been an increase in the number of research and developments in deep learning solutions for object detection applied to driverless vehicles. This application benefited from the growing trend felt in innovative perception solutions, such as LiDAR sensors. Currently, this is the preferred device to accomplish those tasks in autonomous vehicles. There is a broad variety of research works on models based on point clouds, standing out for being efficient and robust in their intended tasks, but they are also characterized by requiring point cloud processing times greater than the minimum required, given the risky nature of the application. This research work aims to provide a design and implementation of a hardware IP optimized for computing convolutions, rectified linear unit (ReLU), padding, and max pooling. This engine was designed to enable the configuration of features such as varying the size of the feature map, filter size, stride, number of inputs, number of filters, and the number of hardware resources required for a specific convolution. Performance results show that by resorting to parallelism and quantization approach, the proposed solution could reduce the amount of logical FPGA resources by 40 to 50%, enhancing the processing time by 50% while maintaining the deep learning operation accuracy.

Application of Hyperspectral Imaging and Deep Learning for Robust Prediction of Sugar and pH Levels in Wine Grape Berries.

Remote sensing technology, such as hyperspectral imaging, in combination with machine learning algorithms, has emerged as a viable tool for rapid and nondestructive assessment of wine grape ripeness. However, the differences in terroir, together with the climatic variations and the variability exhibited by different grape varieties, have a considerable impact on the grape ripening stages within a vintage and between vintages and, consequently, on the robustness of the predictive models. To address this challenge, we present a novel one-dimensional convolutional neural network architecture-based model for the prediction of sugar content and pH, using reflectance hyperspectral data from different vintages. We aimed to evaluate the model's generalization capacity for different varieties and for a different vintage not employed in the training process, using independent test sets. A transfer learning mechanism, based on the proposed convolutional neural network, was also used to evaluate improvements in the model's generalization. Overall, the results for generalization ability showed a very good performance with RMSEP values of 1.118 °Brix and 1.085 °Brix for sugar content and 0.199 and 0.183 for pH, for test sets using different varieties and a different vintage, respectively, improving and updating the current state of the art.

Resource-Constrained Onboard Inference of 3D Object Detection and Localisation in Point Clouds Targeting Self-Driving Applications.

Research about deep learning applied in object detection tasks in LiDAR data has been massively widespread in recent years, achieving notable developments, namely in improving precision and inference speed performances. These improvements have been facilitated by powerful GPU servers, taking advantage of their capacity to train the networks in reasonable periods and their parallel architecture that allows for high performance and real-time inference. However, these features are limited in autonomous driving due to space, power capacity, and inference time constraints, and onboard devices are not as powerful as their counterparts used for training. This paper investigates the use of a deep learning-based method in edge devices for onboard real-time inference that is power-effective and low in terms of space-constrained demand. A methodology is proposed for deploying high-end GPU-specific models in edge devices for onboard inference, consisting of a two-folder flow: study model hyperparameters' implications in meeting application requirements; and compression of the network for meeting the board resource limitations. A hybrid FPGA-CPU board is proposed as an effective onboard inference solution by comparing its performance in the KITTI dataset with computer performances. The achieved accuracy is comparable to the PC-based deep learning method with a plus that it is more effective for real-time inference, power limited and space-constrained purposes.

Real-Time 3D Object Detection and SLAM Fusion in a Low-Cost LiDAR Test Vehicle Setup.

Recently released research about deep learning applications related to perception for autonomous driving focuses heavily on the usage of LiDAR point cloud data as input for the neural networks, highlighting the importance of LiDAR technology in the field of Autonomous Driving (AD). In this sense, a great percentage of the vehicle platforms used to create the datasets released for the development of these neural networks, as well as some AD commercial solutions available on the market, heavily invest in an array of sensors, including a large number of sensors as well as several sensor modalities. However, these costs create a barrier to entry for low-cost solutions for the performance of critical perception tasks such as Object Detection and SLAM. This paper explores current vehicle platforms and proposes a low-cost, LiDAR-based test vehicle platform capable of running critical perception tasks (Object Detection and SLAM) in real time. Additionally, we propose the creation of a deep learning-based inference model for Object Detection deployed in a resource-constrained device, as well as a graph-based SLAM implementation, providing important considerations, explored while taking into account the real-time processing requirement and presenting relevant results demonstrating the usability of the developed work in the context of the proposed low-cost platform.

Radiographic assessment of humeroulnar congruity in a medium and a large breed of dog.

Elbow joint incongruity is recognized as an important factor in the development, treatment, and prognosis of canine elbow dysplasia. Elbow incongruity has been measured based on radiographic joint space widths, however these values can be affected by the degree of elbow joint flexion. Recent studies have reported radiographic curvature radii as more precise measures of humeroulnar congruity in dogs. The aim of this prospective observational study was to describe radiographic curvature radii measured from flexed and extended elbow radiographs for a sample of dogs representing a medium breed (Portuguese Pointing Dog) and a large breed (Estrela Mountain Dog). The curvature radii from the ulnar trochlear notch and humeral trochlea were measured in 114 mediolateral elbow extended radiographic views (30 Portuguese Pointing Dog and 27 Estrela Mountain Dog), and 84 mediolateral flexed views (22 Portuguese Pointing Dog and 20 Estrela Mountain Dog). The sampled animals' ages ranged from 12 to 84 months (34.6 ± 17.8 months). Good agreement was observed between curvature radii measurements for flexed vs. extended views in both breed groups. Ulnar trochlear notch curvature radii measurements were greater than humeral trochlea curvature radii measurements in both breed groups. Both curvature radii were greater in the large-breed dog group vs. the medium-breed dog group. Both breed groups had ulnar and humeral curves with similar typology. However, the large breed group had greater intermediate differences between the humeroulnar surface curvature radii. Results from this study supported the use of curvature radii as measures of humeroulnar congruity in mediolateral flexed elbow radiographs of medium and large breed dogs.

A comparison of two-dimensional and three-dimensional techniques for the determination of hindlimb kinematics during treadmill locomotion in rats following spinal cord injury.

Of all the detrimental effects of spinal cord injury (SCI), one of the most devastating effects is the disruption of the ability to walk. Therefore, much effort has been focused on developing several methods to document the recovery of locomotor function after experimental SCI. Computerized rat gait analysis is becoming increasingly popular in the SCI research community. The two-dimensional (2D) kinematic approach is by far the most popular technique in rat gait analysis. This is a simple inexpensive procedure, which requires only one camera to record the movement. Our study included an examination of locomotion on a treadmill using 2D and three-dimensional (3D) analysis, in neurologically intact animals and following moderate T9 contusion injury. Despite the overall time course patterns of the curves were identical, we found significant differences between values of the 2D and 3D joint angular motion. In conclusion, maximal precision and accuracy of the kinematic values are expected when the experimental protocol includes a 3D motion analysis methodology. Moreover, a 2D method cannot be used to determine the external or internal rotations of the foot because this movement occurs in the transverse plane.

Passive hip laxity in Estrela Mountain Dog--distraction index, heritability and breeding values.

Two hundred and fifteen Estrela Mountain Dogs (EMD) were examined using the PennHIP method between 2002 and 2006. Passive hip laxity (PHL) was estimated calculating the distraction index (DI). Pedigree information was obtained from the Portuguese Kennel Club. The heritability and breeding values were estimated using the linear Animal Model. The DI, using as reference the worst joint of each animal, ranged from 0.15 to 1.12 (0.55 +/- 0.19). The PHL was equal in right and left sides, and was higher in females than in males (P > 0.05 and P <0.01 in the t-test, respectively). The estimated heritability was 0.83 +/- 0.11. The mean breeding values for PHL were stable in EMD born between 1991 and 2003, and showed an improvement in 2004 and 2005. The data confirm high PHL in breeds with high prevalence and severity of canine hip dysplasia. The high heritability indicates that the DI could be reduced in the breed if PHL were used as a selection criterion. The early favourable evolution of DI breeding values could be related with the increase of voluntary radiographic hip screening in the last years, and the subsequent introduction of hip quality as a breeding criterion.

Computed tomography and radiographic assessment of congruity between the ulnar trochlear notch and humeral trochlea in large breed dogs.

OBJECTIVE: Assess the fit between the ulnar trochlear notch (UTN) and humeral trochlea (HT) in elbow radiographs and computed tomography (CT) images by measuring the curvature radii in normal joints. METHODS: Mediolateral extended view (MLE) and CT were performed on 16 elbows from cadaveric dogs weighing over 20 kg. Curves were traced at the subchondral level from the UTN central ridge and the HT sagittal groove related to the UTN on ~132° joint extension. Curvature radii were achieved using appropriate computer software. RESULTS: Intraclass correlation coefficient between MLE/CT measurements were statistically significant, with lower limits of a 95% confidence interval (CI) >0.75. The mean differences between MLE/CT measurements were -0.71 mm (95% CI: -0.97 to -0.45) for UTN and -1.04 mm (95% CI: -1.21 to -0.87) for HT. The UTN and HT curvature radii typology were similar, with maximum radius values at the starting point and two intermediate peaks. The UTN curvature radii were bigger than the HT radii, with the largest differences in the most proximal aspect of the joint and in the second intermediate peak. CLINICAL SIGNIFICANCE: To the authors' knowledge, this is the first report on humeroulnar congruity through curvature radii evaluation. The software and methodology proposed enabled an adequate UTN and HT radii curvature assessment in MLE and CT images. Comparative studies in normal and dysplastic dogs of various breeds could better clarify the role of UTN and HT curvature radii in elbow incongruity and dysplasia.

Characterization of neural network generalization in the determination of pH and anthocyanin content of wine grape in new vintages and varieties.

The generalization ability of hyperspectral imaging combined with neural networks (NN) in estimating pH and anthocyanin content during ripening was evaluated for vintages and varieties not employed in the NN creation. A NN, from a previously published work, trained with grape samples of Touriga Franca (TF) variety harvested in 2012 was tested with TF from 2013 and two new varieties, Touriga Nacional (TN) and Tinta Barroca (TB) from 2013. Each sample contained a small number of whole berries. The present work results suggest that, under certain conditions, it might be possible for the NN to provide for new vintages and varieties results comparable to those of the vintages and varieties employed in the NN training. For pH, the results are state-of-the-art for the new vintage and varieties tested. For anthocyanin, generalization is bad for TB from 2013 but presents state-of-the-art absolute percentage error for TF and TN from 2013.

Effect of skin movement on the analysis of hindlimb kinematics during treadmill locomotion in rats.

In rat gait kinematics, the method most frequently used for measuring hindlimb movement involves placing markers on the skin surface overlying the joints being analyzed. Soft tissue movement around the knee joint has been considered the principle source of error when estimating hindlimb joint kinematics in rodents. However, the motion of knee marker was never quantified, nor the different variations in joint angle associated with this gait analysis system. The purpose of this study was two-fold. The first purpose was to expand upon the limited pool of information describing the effect of soft tissue movement over the knee upon the angular positions of the hip, knee and ankle of rats during treadmill locomotion. Secondly, it was a goal of this study to document the magnitude of the skin displacement when using markers that were attached superficially to the knee joint. This was examined by comparing the hindlimb kinematics in sagittal plane during treadmill locomotion determined from the marker attached to the knee and when the knee position was determined indirectly by computer analysis. Results showed that there is a considerable skin movement artefact which propagates to knee joint position and hindlimb kinematics estimates. It was concluded that these large errors can decrease data reliability in the research of rat gait analysis.

A comparison analysis of hindlimb kinematics during overground and treadmill locomotion in rats.

The convenience of the motor-driven treadmill makes it an attractive instrument for investigating rat locomotion. However, no data are available to indicate whether hindlimb treadmill kinematic findings may be compared or generalized to overground locomotion. In this investigation, we compared overground and treadmill locomotion for differences in the two-dimensional angular kinematics and temporal and spatial measurements for the hindlimb. Ten female rats were evaluated at the same speed for natural overground and treadmill walking. The walking velocity, swing duration and stride length were statistically indistinguishable between the two testing conditions. Significant differences were found between overground and treadmill locomotion for step cycle duration and stance phase duration parameters. During the stance phase of walking, the angular movement of the hip, knee and ankle joints were significantly different in the two conditions, with greater flexion occurring on the overground. Despite this, the sagittal joint movements of the hindlimb were similar between the two walking conditions, with only three parameters being significantly different in the swing. Hip height and angle-angle cyclograms were also only found to display subtle differences. This study suggests that reliable kinematic measurements can be obtained from the treadmill gait analysis in rats.

Use of Visible and Short-Wave Near-Infrared Hyperspectral Imaging To Fingerprint Anthocyanins in Intact Grape Berries.

In red grape berries, anthocyanins account for about 50% of the skin phenols and are responsible for the final wine color. Individual anthocyanin levels and compositional profiles vary with cultivar, maturity, season, region, and yield and have been proposed as chemical markers to differentiate wines and to provide valuable information regarding the adulteration of musts and wines. A fast, easy, solvent-free, nondestructive method based on visible, short-wave, and near-infrared hyperspectral imaging (HSI) in intact grape berries to fingerprint the color pigments in eight different grape varieties was developed and tested against HPLC. Predictive models based on modified partial least-squares (MPLS) were built for 14 individual anthocyanins with coefficients of determination of cross-validation (R2CV) ranging from 0.70 to 0.93. For the grouping of total and nonacylated anthocyanins, external validation was conducted with coefficient of determination of prediction (R2P) of 0.86. HSI could potentially become an alternative to HPLC with reduced analysis time and labor costs while providing reliable and robust information on the anthocyanin composition of grape berries.

Curvature Radius Measurements From the Ulnar Trochlear Notch in Large Dogs.

Assessing the ulnar trochlear notch (UTN) radiographic anatomy has been considered important, but difficult, in the diagnosis of elbow dysplasia. The purpose of this study was to evaluate UTN curvature of natural elbows in radiographs, using a methodology applied to disarticulated joints. The methodology was implemented and validated using dedicated software created by the authors. Mediolateral extended (MLE) and mediolateral flexed (MLF) elbow views were used from 20 joints from canine cadavers that were over 20 kg. After arranging the bones to avoid radiographic overlapping of the bones, an additional mediolateral radioulnar (MLRU) view was made. Curvature radius measurements from the central ridge of the UTN of each elbow were acquired in the MLRU view, using the software. The measurements were repeated in a second session, to determine repeatability. Then similar UTN measurements were taken from the MLE and MLF views, to determine reproducibility. Intraclass correlation coefficient (ICC) for repeatability and reproducibility of measurements were above 0.98 (95% confidence interval limits >0.75). The 95% limits of agreement (LA) for repeatability were -2.98 to 3.19 mm. The 95% LA for reproducibility between MLRU and MLE views were -4.32 to 3.75 mm. The 95% LA for reproducibility between MLRU and MLF views were -5.02 to 4.07 mm. The methodology and software are determined to be both precise and suitable to evaluate the UTN in MLE and MLF elbow views of large breed dogs, for anatomical and clinical purposes. In the future it would be useful to characterize normal and dysplastic UTN of different dog breeds.

Functional and morphological assessment of a standardized rat sciatic nerve crush injury with a non-serrated clamp.

Peripheral nerve researchers frequently use the rat sciatic nerve crush as a model for axonotmesis. Unfortunately, studies from various research groups report results from different crush techniques and by using a variety of evaluation tools, making comparisons between studies difficult. The purpose of this investigation was to determine the sequence of functional and morphologic changes after an acute sciatic nerve crush injury with a non-serrated clamp, giving a final standardized pressure of p = 9 MPa. Functional recovery was evaluated using the sciatic functional index (SFI), the extensor postural thrust (EPT) and the withdrawal reflex latency (WRL), before injury, and then at weekly intervals until week 8 postoperatively. The rats were also evaluated preoperatively and at weeks 2, 4, and 8 by ankle kinematics, toe out angle (TOA), and gait-stance duration. In addition, the motor nerve conduction velocity (MNCV) and the gastrocnemius-soleus weight parameters were measured just before euthanasia. Finally, structural, ultrastructural and histomorphometric analyses were carried out on regenerated nerve fibers. At 8 weeks after the crush injury, a full functional recovery was predicted by SFI, EPT, TOA, and gait-stance duration, while all the other parameters were still recovering their original values. On the other hand, only two of the histomorphometric parameters of regenerated nerve fibers, namely myelin thickness/axon diameter ratio and fiber/axon diameter ratio, returned to normal values while all other parameters were significantly different from normal values. The employment of traditional methods of functional evaluation in conjunction with the modern techniques of computerized analysis of gait and histomorphometric analysis should thus be recommended for an overall assessment of recovery in the rat sciatic nerve crush model.

Methods for the experimental functional assessment of rat sciatic nerve regeneration.

In experimental peripheral nerve studies, the rat sciatic nerve model is widely used to examine functional changes after different surgical repairs or pharmacological treatments, following nerve injury. The number and diversity of tests which have been used to assess functional recovery after experimental interventions often makes it difficult to recommend any particular indicator of nerve regeneration. Functional assessment after sciatic nerve lesion has long been focused on walking track analysis, therefore, this article describes in more detail the method to obtain and measure the walking tracks in order to calculate the sciatic functional index (SFI). However, it is important to note that the validity of the SFI has been questioned by several researchers. In addition, the present review includes other traditional tests described in the experimental peripheral nerve literature regarding the rate of return of motor function and sensation, such as the extensor postural thrust (EPT), nociceptive function, and the gastrocnemius-soleus weight parameters. In the last decade, several authors have designed a series of sensitive quantitative methods to assess the recovery of hind limb locomotor function using computerized rat gait analysis. This study aims to review kinematic measures that can be gathered with this technology, including calculation of sciatic functional index, gait-stance duration, ankle kinematics and toe out angle (TOA). A combination of tests, each examining particular components of recovered sensorimotor function is recommended for an overall assessment of rat sciatic nerve regeneration.

Automatic histogram threshold using fuzzy measures.

In this paper, an automatic histogram threshold approach based on a fuzziness measure is presented. This work is an improvement of an existing method. Using fuzzy logic concepts, the problems involved in finding the minimum of a criterion function are avoided. Similarity between gray levels is the key to find an optimal threshold. Two initial regions of gray levels, located at the boundaries of the histogram, are defined. Then, using an index of fuzziness, a similarity process is started to find the threshold point. A significant contrast between objects and background is assumed. Previous histogram equalization is used in small contrast images. No prior knowledge of the image is required.

Comparison of clinical, radiographic, computed tomographic, and magnetic resonance imaging methods for early prediction of canine hip laxity and dysplasia.

The purpose of the study was to use two palpation methods (Bardens and Ortolani), a radiographic distraction view, three computed tomography (CT) measurements (dorsolateral subluxation score, the lateral center-edge angle, and acetabular ventroversion angle) and two magnetic resonance (MR) imaging hip studies (synovial fluid and acetabular depth indices) in the early monitoring of hip morphology and laxity in 7-9 week old puppies; and in a follow-up study to compare their accuracy in predicting later hip laxity and dysplasia. The MR imaging study was performed with the dog in dorsal recumbency and the CT study with the animal in a weight-bearing position. There was no association between clinical laxity with later hip laxity or dysplasia. The dorsolateral subluxation score and the lateral center-edge angle were characterized by a weak negative correlation with later radiographic passive hip laxity (-0.26 < r < -0.38, P < 0.05) but its association with hip dysplasia was not significant. There was an association between early radiographic passive hip laxity and synovial fluid index with later passive hip laxity (0.41 < r < 0.55, P < 0.05) and this was significantly different in dysplastic vs. nondysplastic hips (P < 0.05). There was no association between the remaining variables and later hip laxity or dysplasia. The overlapping ranges of early passive hip laxity and synovial fluid index for hip dysplasia grades and the moderate correlations with the later passive hip laxity make the results of these variables unreliable for use in predicting hip laxity and dysplasia susceptibility.

Dynamic joint stiffness of the ankle during walking: gender-related differences.

OBJECTIVES: To characterize and compare dynamic joint stiffness (DJS) of the ankle in the sagittal plane during natural cadence walking in both genders. DESIGN: Observation, cross-sectional and matched pairs. PARTICIPANTS: Twenty-one males (mean age=27+/-4.2 years) and 18 females (mean age=22.9+/-4.1 years). SETTING: Laboratory at the University de Trás-os-Montes and Alto Douro. The DJS of the ankle was assessed while participants performed barefoot walking at a natural cadence, was measured using a force platform (1000 Hz), and synchronized with a high-speed camera (200 Hz). MAIN OUTCOME MEASURES: DJS of the ankle in males and females was calculated in two sub-phases (the second and the fourth sub-phases) of the stance phase. The Wilcoxon non-parametric matched-pairs test and the Mann-Whitney U non-parametric test were used to analyse difference of DJS of the ankle within and between the groups with p0.05 regarded as significant. RESULTS: Male and female participants demonstrated significantly (p<0.005) less DJS during the second sub-phase. DJS was not significantly (p>0.063) different between females (0.0511 N m/kg/ degrees ) and males (0.0596 N m/kg/ degrees ) during the second sub-phase. DJS was significantly (p<0.001) higher in males (0.0844 N m/kg/ degrees ) than females (0.0691 N m/kg/ degrees ) during the fourth sub-phase. CONCLUSIONS: Gender-specific DJS characteristics demonstrated by females in this study may be used in preventative training programs designed to promote a greater ability to use contractile components to produce mechanical energy through the ankle.