Closing the Wearable Gap: Foot-ankle kinematic modeling via deep learning models based on a smart sock wearable.

The development of wearable technology, which enables motion tracking analysis for human movement outside the laboratory, can improve awareness of personal health and performance. This study used a wearable smart sock prototype to track foot-ankle kinematics during gait movement. Multivariable linear regression and two deep learning models, including long short-term memory (LSTM) and convolutional neural networks, were trained to estimate the joint angles in sagittal and frontal planes measured by an optical motion capture system. Participant-specific models were established for ten healthy subjects walking on a treadmill. The prototype was tested at various walking speeds to assess its ability to track movements for multiple speeds and generalize models for estimating joint angles in sagittal and frontal planes. LSTM outperformed other models with lower mean absolute error (MAE), lower root mean squared error, and higher R-squared values. The average MAE score was less than 1.138° and 0.939° in sagittal and frontal planes, respectively, when training models for each speed and 2.15° and 1.14° when trained and evaluated for all speeds. These results indicate wearable smart socks to generalize foot-ankle kinematics over various walking speeds with relatively low error and could consequently be used to measure gait parameters without the need for a lab-constricted motion capture system.

A Survey on Deep-Learning-Based LiDAR 3D Object Detection for Autonomous Driving.

LiDAR is a commonly used sensor for autonomous driving to make accurate, robust, and fast decision-making when driving. The sensor is used in the perception system, especially object detection, to understand the driving environment. Although 2D object detection has succeeded during the deep-learning era, the lack of depth information limits understanding of the driving environment and object location. Three-dimensional sensors, such as LiDAR, give 3D information about the surrounding environment, which is essential for a 3D perception system. Despite the attention of the computer vision community to 3D object detection due to multiple applications in robotics and autonomous driving, there are challenges, such as scale change, sparsity, uneven distribution of LiDAR data, and occlusions. Different representations of LiDAR data and methods to minimize the effect of the sparsity of LiDAR data have been proposed. This survey presents the LiDAR-based 3D object detection and feature-extraction techniques for LiDAR data. The 3D coordinate systems differ in camera and LiDAR-based datasets and methods. Therefore, the commonly used 3D coordinate systems are summarized. Then, state-of-the-art LiDAR-based 3D object-detection methods are reviewed with a selected comparison among methods.

Comparison of the Capacitance of a Cyclically Fatigued Stretch Sensor to a Non-Fatigued Stretch Sensor When Performing Static and Dynamic Foot-Ankle Motions.

Motion capture is the current gold standard for assessing movement of the human body, but laboratory settings do not always mimic the natural terrains and movements encountered by humans. To overcome such limitations, a smart sock that is equipped with stretch sensors is being developed to record movement data outside of the laboratory. For the smart sock stretch sensors to provide valuable feedback, the sensors should have durability of both materials and signal. To test the durability of the stretch sensors, the sensors were exposed to high-cycle fatigue testing with simultaneous capture of the capacitance. Following randomization, either the fatigued sensor or an unfatigued sensor was placed in the plantarflexion position on the smart sock, and participants were asked to complete the following static movements: dorsiflexion, inversion, eversion, and plantarflexion. Participants were then asked to complete gait trials. The sensor was then exchanged for either an unfatigued or fatigued plantarflexion sensor, depending upon which sensor the trials began with, and each trial was repeated by the participant using the opposite sensor. Results of the tests show that for both the static and dynamic movements, the capacitive output of the fatigued sensor was consistently higher than that of the unfatigued sensor suggesting that an upwards drift of the capacitance was occurring in the fatigued sensors. More research is needed to determine whether stretch sensors should be pre-stretched prior to data collection, and to also determine whether the drift stabilizes once the cyclic softening of the materials comprising the sensor has stabilized.

Class-Aware Fish Species Recognition Using Deep Learning for an Imbalanced Dataset.

Fish species recognition is crucial to identifying the abundance of fish species in a specific area, controlling production management, and monitoring the ecosystem, especially identifying the endangered species, which makes accurate fish species recognition essential. In this work, the fish species recognition problem is formulated as an object detection model to handle multiple fish in a single image, which is challenging to classify using a simple classification network. The proposed model consists of MobileNetv3-large and VGG16 backbone networks and an SSD detection head. Moreover, a class-aware loss function is proposed to solve the class imbalance problem of our dataset. The class-aware loss takes the number of instances in each species into account and gives more weight to those species with a smaller number of instances. This loss function can be applied to any classification or object detection task with an imbalanced dataset. The experimental result on the large-scale reef fish dataset, SEAMAPD21, shows that the class-aware loss improves the model over the original loss by up to 79.7%. The experimental result on the Pascal VOC dataset also shows the model outperforms the original SSD object detection model.

Off-Road Detection Analysis for Autonomous Ground Vehicles: A Review.

When it comes to some essential abilities of autonomous ground vehicles (AGV), detection is one of them. In order to safely navigate through any known or unknown environment, AGV must be able to detect important elements on the path. Detection is applicable both on-road and off-road, but they are much different in each environment. The key elements of any environment that AGV must identify are the drivable pathway and whether there are any obstacles around it. Many works have been published focusing on different detection components in various ways. In this paper, a survey of the most recent advancements in AGV detection methods that are intended specifically for the off-road environment has been presented. For this, we divided the literature into three major groups: drivable ground and positive and negative obstacles. Each detection portion has been further divided into multiple categories based on the technology used, for example, single sensor-based, multiple sensor-based, and how the data has been analyzed. Furthermore, it has added critical findings in detection technology, challenges associated with detection and off-road environment, and possible future directions. Authors believe this work will help the reader in finding literature who are doing similar works.

WCNN3D: Wavelet Convolutional Neural Network-Based 3D Object Detection for Autonomous Driving.

Three-dimensional object detection is crucial for autonomous driving to understand the driving environment. Since the pooling operation causes information loss in the standard CNN, we designed a wavelet-multiresolution-analysis-based 3D object detection network without a pooling operation. Additionally, instead of using a single filter like the standard convolution, we used the lower-frequency and higher-frequency coefficients as a filter. These filters capture more relevant parts than a single filter, enlarging the receptive field. The model comprises a discrete wavelet transform (DWT) and an inverse wavelet transform (IWT) with skip connections to encourage feature reuse for contrasting and expanding layers. The IWT enriches the feature representation by fully recovering the lost details during the downsampling operation. Element-wise summation was used for the skip connections to decrease the computational burden. We trained the model for the Haar and Daubechies (Db4) wavelets. The two-level wavelet decomposition result shows that we can build a lightweight model without losing significant performance. The experimental results on KITTI's BEV and 3D evaluation benchmark show that our model outperforms the PointPillars-based model by up to 14% while reducing the number of trainable parameters.

Fatigue Testing of Wearable Sensing Technologies: Issues and Opportunities.

Standards for the fatigue testing of wearable sensing technologies are lacking. The majority of published fatigue tests for wearable sensors are performed on proof-of-concept stretch sensors fabricated from a variety of materials. Due to their flexibility and stretchability, polymers are often used in the fabrication of wearable sensors. Other materials, including textiles, carbon nanotubes, graphene, and conductive metals or inks, may be used in conjunction with polymers to fabricate wearable sensors. Depending on the combination of the materials used, the fatigue behaviors of wearable sensors can vary. Additionally, fatigue testing methodologies for the sensors also vary, with most tests focusing only on the low-cycle fatigue (LCF) regime, and few sensors are cycled until failure or runout are achieved. Fatigue life predictions of wearable sensors are also lacking. These issues make direct comparisons of wearable sensors difficult. To facilitate direct comparisons of wearable sensors and to move proof-of-concept sensors from "bench to bedside", fatigue testing standards should be established. Further, both high-cycle fatigue (HCF) and failure data are needed to determine the appropriateness in the use, modification, development, and validation of fatigue life prediction models and to further the understanding of how cracks initiate and propagate in wearable sensing technologies.

Wearable Stretch Sensors for Human Movement Monitoring and Fall Detection in Ergonomics.

Wearable sensors are beneficial for continuous health monitoring, movement analysis, rehabilitation, evaluation of human performance, and for fall detection. Wearable stretch sensors are increasingly being used for human movement monitoring. Additionally, falls are one of the leading causes of both fatal and nonfatal injuries in the workplace. The use of wearable technology in the workplace could be a successful solution for human movement monitoring and fall detection, especially for high fall-risk occupations. This paper provides an in-depth review of different wearable stretch sensors and summarizes the need for wearable technology in the field of ergonomics and the current wearable devices used for fall detection. Additionally, the paper proposes the use of soft-robotic-stretch (SRS) sensors for human movement monitoring and fall detection. This paper also recapitulates the findings of a series of five published manuscripts from ongoing research that are published as Parts I to V of "Closing the Wearable Gap" journal articles that discuss the design and development of a foot and ankle wearable device using SRS sensors that can be used for fall detection. The use of SRS sensors in fall detection, its current limitations, and challenges for adoption in human factors and ergonomics are also discussed.

Closing the Wearable Gap-Part II: Sensor Orientation and Placement for Foot and Ankle Joint Kinematic Measurements.

The linearity of soft robotic sensors (SRS) was recently validated for movement angle assessment using a rigid body structure that accurately depicted critical movements of the foot-ankle complex. The purpose of this study was to continue the validation of SRS for joint angle movement capture on 10 participants (five male and five female) performing ankle movements in a non-weight bearing, high-seated, sitting position. The four basic ankle movements-plantar flexion (PF), dorsiflexion (DF), inversion (INV), and eversion (EVR)-were assessed individually in order to select good placement and orientation configurations (POCs) for four SRS positioned to capture each movement type. PF, INV, and EVR each had three POCs identified based on bony landmarks of the foot and ankle while the DF location was only tested for one POC. Each participant wore a specialized compression sock where the SRS could be consistently tested from all POCs for each participant. The movement data collected from each sensor was then compared against 3D motion capture data. R-squared and root-mean-squared error averages were used to assess relative and absolute measures of fit to motion capture output. Participant robustness, opposing movements, and gender were also used to identify good SRS POC placement for foot-ankle movement capture.

Semantic Segmentation with Transfer Learning for Off-Road Autonomous Driving.

Since the state-of-the-art deep learning algorithms demand a large training dataset, which is often unavailable in some domains, the transfer of knowledge from one domain to another has been a trending technique in the computer vision field. However, this method may not be a straight-forward task considering several issues such as original network size or large differences between the source and target domain. In this paper, we perform transfer learning for semantic segmentation of off-road driving environments using a pre-trained segmentation network called DeconvNet. We explore and verify two important aspects regarding transfer learning. First, since the original network size was very large and did not perform well for our application, we proposed a smaller network, which we call the light-weight network. This light-weight network is half the size to the original DeconvNet architecture. We transferred the knowledge from the pre-trained DeconvNet to our light-weight network and fine-tuned it. Second, we used synthetic datasets as the intermediate domain before training with the real-world off-road driving data. Fine-tuning the model trained with the synthetic dataset that simulates the off-road driving environment provides more accurate results for the segmentation of real-world off-road driving environments than transfer learning without using a synthetic dataset does, as long as the synthetic dataset is generated considering real-world variations. We also explore the issue whereby the use of a too simple and/or too random synthetic dataset results in negative transfer. We consider the Freiburg Forest dataset as a real-world off-road driving dataset.

Fusion of an Ensemble of Augmented Image Detectors for Robust Object Detection.

A significant challenge in object detection is accurate identification of an object's position in image space, whereas one algorithm with one set of parameters is usually not enough, and the fusion of multiple algorithms and/or parameters can lead to more robust results. Herein, a new computational intelligence fusion approach based on the dynamic analysis of agreement among object detection outputs is proposed. Furthermore, we propose an online versus just in training image augmentation strategy. Experiments comparing the results both with and without fusion are presented. We demonstrate that the augmented and fused combination results are the best, with respect to higher accuracy rates and reduction of outlier influences. The approach is demonstrated in the context of cone, pedestrian and box detection for Advanced Driver Assistance Systems (ADAS) applications.

Pullout strength of monocortical and bicortical screws in metaphyseal and diaphyseal regions of the canine humerus.

OBJECTIVE: Monocortical screws are commonly employed in locking plate fixation, but specific recommendations for their placement are lacking and use of short monocortical screws in metaphyseal bone may be contraindicated. Objectives of this study were to evaluate axial pullout strength of two different lengths of monocortical screws placed in various regions of the canine humerus compared to bicortical screws, and to derive cortical thickness and bone density values for those regions using quantitative computed tomography analysis (QCT). METHODS: The QCT analysis was performed on 36 cadaveric canine humeri for six regions of interest (ROI). A bicortical, short monocortical, or 50% transcortical 3.5 mm screw was implanted in each ROI and axial pullout testing was performed. RESULTS: Bicortical screws were stronger than monocortical screws in all ROI except the lateral epicondylar crest. Short monocortical metaphyseal screws were weaker than those placed in other regions. The 50% transcortical screws were stronger than the short monocortical screws in the condyle. A linear relationship between screw length and pullout strength was observed. CLINICAL SIGNIFICANCE: Cortical thickness and bone density measurements were obtained from multiple regions of the canine humerus using QCT. Use of short monocortical screws may contribute to failure of locking plate fixation of humeral fractures, especially when placed in the condyle. When bicortical screw placement is not possible, maximizing monocortical screw length may optimize fixation stability for distal humeral fractures.

A multi-classifier and decision fusion framework for robust classification of mammographic masses.

Most end-to-end Computer Aided Diagnosis (CAD) systems follow a three step approach - (1) Image enhancement and segmentation, (2) Feature extraction, and, (3) Classification. While the state of the art in image enhancement and segmentation can now very accurately identify regions of interest for feature extraction, they typically result in very high dimensional feature spaces. This adversely affects the performance of classification systems because a large feature space dimensionality necessitates a large training database to accurately model the statistics of class features (e.g. benign versus malignant classes). In this work, we present a robust multi-classifier decision fusion framework that employs a divide-and-conquer approach for alleviating the affects of high dimensionality of feature vectors. The feature space is partitioned into multiple smaller sized spaces, and a bank of classifiers (a multi-classifier system) is employed to perform classification in each of the partition. Finally, a decision fusion system merges decisions from each classifier in the bank into a single decision. The system is applied to the problem of classifying digital mammographic masses as either benign or malignant.

Level set-based core segmentation of mammographic masses facilitating three stage (core, periphery, spiculation) analysis.

We present mammographic mass core segmentation, based on the Chan-Vese level set method. The proposed method is analyzed via resulting feature efficacies. Additionally, the core segmentation method is used to investigate the idea of a three stage segmentation approach, i.e. segment the mass core, periphery, and spiculations (if any exist) and use features from these three segmentations to classify the mass as either benign or malignant. The proposed core segmentation method and a proposed end-to-end computer aided detection (CAD) system using a three stage segmentation are implemented and experimentally tested with a set of 60 mammographic images from the Digital Database of Screening Mammography. Receiver operating characteristic (ROC) curve AZ values for morphological and texture features extracted from the core segmentation are shown to be on par, or better, than those extracted from a periphery segmentation. The efficacy of the core segmentation features when combined with the periphery and spiculation segmentation features are shown to be feature set dependent. The proposed end-to-end system uses stepwise linear discriminant analysis for feature selection and a maximum likelihood classifier. Using all three stages (core + periphery + spiculations) results in an overall accuracy (OA) of 90% with 2 false negatives (FN). Since many CAD systems only perform a periphery analysis, adding core features could be a benefit to potentially increase OA and reduce FN cases.

Digital mammographic computer aided diagnosis (CAD) using adaptive level set segmentation.

We present a mammographic computer aided diagnosis (CAD) system, which uses an adaptive level set segmentation method (ALSSM), which segments suspicious masses in the polar domain and adaptively adjusts the border threshold at each angle to provide high-quality segmentation results. The primary contribution of this paper is the adaptive speed function for controlling level set segmentation. To assess the efficacy of the system, 60 relatively difficult cases (30 benign, 30 malignant) from the Digital Database of Screening Mammography (DDSM) are analyzed. The segmentation efficacy is analyzed qualitatively via visual inspection and quantitatively via the area under the receiver operating characteristics (ROC) curve (AZ) and classification accuracies. For the ALSSM, the best results are 87% overall accuracy, A(Z)=0.9687 with 28/30 malignant cases detected. The qualitative and quantitative results show that the ALSSM provides excellent segmentation and classification results and compares favorably to previous CAD systems in the literature which also used the DDSM database.

Digital mammogram spiculated mass detection and spicule segmentation using level sets.

This letter presents an automated mammographic computer aided diagnosis (CAD) system to detect and segment spicules in digital mammograms, termed spiculation segmentation with level sets (SSLS). SSLS begins with a segmentation of the suspicious mass periphery, which is created using a previously developed adaptive level set segmentation algorithm (ALSSM) by the authors. The mammogram is then analyzed using features derived from the Dixon and Taylor Line Operator (DTLO), which is a method of linear structure enhancement. Features are extracted, optimized, and then the suspicious mass is classified as benign or malignant. To assess the system efficacy, 60 difficult mammographic images from the Digital Database of Screening Mammography (DDSM), containing 30 benign non-spiculated cases, 17 malignant spiculated cases, and 13 malignant non-spiculated cases, are analyzed. The initial spiculation detection method found 100% of the spiculated lesions with no false positive detections, and has area under the receiver operating characteristics (ROC) curve A(Z)=1.0. The values using ALSSM (periphery segmentation only) are A(Z)=0.9687 and 0.9708 for two investigated feature sets, and increases to A(Z)=0.986 2 using SSLS (spiculation segmentation). The best classification results are 93% overall accuracy (OA), with three false positives (FP) and one false negative (FN) using a 1-NN (Nearest Neighbor) or 2-NN classifier, and 92% OA with three FP and two FN using a maximum likelihood classifier.