3D Laser Triangulation and Deep Learning Approach to Tunnel Inspection

Regular inspection of tunnel surfaces is an important practice from both a safety and tunnel asset management perspective. However, inspection for cracking and spalling is still predominantly a manual task, which is time consuming, subjective, and exposes on-foot staff to risk. This presentation will explore the use of 3D laser scanning technology and artificial intelligence to automate the inspection process with a Canadian metro case study being presented.

Terrain surveillance system with drone and applied machine vision

Road accidents are a major contribution to the Annual death rates all over the world. India, ranks first globally in the number of fatalities from road accidents. According to the Ministry of Roads & Transportation, India saw over 440,000 road accidents in 2019. As a result, over 150,000 lives were lost. Poor road conditions contribute to these directly and indirectly. In India, safety standards and conditions of roads are maintained by local bodies in a given area of jurisdiction. While there have been several attempts at improving the quality of roads, weren’t instrumental in giving proper results [42]. A recent study suggested that Artificial Intelligence (AI) might help achieve the goals. Some of the AI applications have had better results when powered with Computer Vision. While computer vision has been previously used to identify faults in roads, it is not widely implemented or made available for public use. Road inspection still largely remains a time-consuming manual task, hindering the maintenance process in most cities. Moreover, being unaware of unattended faults on roads is often the cause of road accidents, especially in rough weather conditions that make it impossible for drivers to visually gauge any dangers on their route. The proposed model uses a transfer-learning approach; using Mask R-CNN in identifying the defects at an instance level segmentation. As adding this, it requires less labelling and an additional mask helps in blocking out extra noise around the images. This paper trains a Mask R-CNN architecture-based model to identify potholes, discontinuous roads, blind spots, speed bumps, and the type of road--gravel, concrete, asphalt, tar, or mud--with a dataset of images obtained from a drone. The model is further trained to create depth maps and friction estimates of the roads being surveyed. Once trained, the model is tested on a drone-captured live feed of roads in Chennai, India. The results, once sufficiently accurate, will be implemented in a practical application to help users assess road conditions on their path.

The effects of a secondary task on gait in axial spondyloarthritis

Studies on the effects of dual tasking in patients with chronic inflammatory rheumatic diseases are limited. The aim of this study was to assess dual tasking while walking in patients with axial spondyloarthritis (axSpA) in comparison to healthy controls. Thirty patients with axSpA and thirty healthy controls underwent a 10-m walk test at a self-selected comfortable walking speed in single- and dual-task conditions. Foot-worn inertial sensors were used to compute spatiotemporal gait parameters. Analysis of spatiotemporal gait parameters showed that the secondary manual task negatively affected walking performance in terms of significantly decreased mean speed (p < 0.001), stride length (p < 0.001) and swing time (p = 0.008) and increased double support (p = 0.002) and stance time (p = 0.008). No significant interaction of group and condition was observed. Both groups showed lower gait performance in dual task condition by reducing speed, swing time and stride length, and increasing double support and stance time. Patients with axSpA were not more affected by the dual task than matched healthy controls, suggesting that the secondary manual task did not require greater attention in patients with axSpA. Increasing the complexity of the walking and/or secondary task may increase the sensitivity of the dual-task design to axial spondyloarthritis.

Route Generation Heuristics for the Automated Creation of Hot Spot Policing Patrol Routes

Hot spot policing is a form of targeted police patrol deployment for decreasing crime. For hot spot policing to be effective, it requires analysis of crime data to identify the specific locations where crime is concentrated and create suitable patrol routes. The creation of hot spot policing patrol routes is a manual task that police officers perform, requiring skills and knowledge about hot spot policing and crime pattern analysis. This can limit the use of hot spot policing where these skills and knowledge are not available, and where they are available, the creation of patrol routes can be a time-consuming task. In this paper, we introduce two computational route generation heuristics that automate creating hot spot policing patrol routes. Both approaches identify the specific locations where crime concentrates and then use different methods to create the patrol routes. We compare the performance of each approach using metrics associated with effective patrol route creation and through visual inspection. We conclude that the heuristics we introduce provide an accurate means for creating hot spot policing patrol routes, which can support greater and improved use of hot spot policing as an effective type of intervention for decreasing crime.

Convolutional neural network for semantic segmentation of fetal echocardiography based on four-chamber view

The acute shortage of trained and experienced sonographers causes the detection of congenital heart defects (CHDs) extremely difficult. In order to minimize this difficulty, an accurate fetal heart segmentation to the early location of such structural heart abnormalities prior to delivery is essential. However, the segmentation process is not an easy task due to the small size of the fetal heart structure. Moreover, the manual task for identifying the standard cardiac planes, primarily based on a four-chamber view, requires a well-trained clinician and experience. In this paper, a CNN method using U-Net architecture was proposed to automate fetal cardiac standard planes segmentation from ultrasound images. A total of 519 fetal cardiac images was obtained from three videos. All data is divided into training and testing data. The testing data consist of 106 slices of the four-chamber segmentation tasks, i.e. atrial septal defect (ASD), ventricular septal defect (VSD), and normal. The segmentation of the post-processing method is needed to enhanced the segmentation result. In this paper, a combination technique with U-Net and Otsu thresholding gives the best performances with 99.48%-pixel accuracy, 96.73% mean accuracy, 94.92% mean intersection over union, and 0.21% segmentation error. In the future, the implementation of Deep Learning in the study of CHDs holds significant potential for identifying novel CHDs in heterogeneous fetal hearts.

Gait Variability and Complexity during Single and Dual-Task Walking on Different Surfaces in Outdoor Environment

Nowadays, gait assessment in the real life environment is gaining more attention. Therefore, it is desirable to know how some factors, such as surfaces (natural, artificial) or dual-tasking, influence real life gait pattern. The aim of this study was to assess gait variability and gait complexity during single and dual-task walking on different surfaces in an outdoor environment. Twenty-nine healthy young adults aged 23.31 ± 2.26 years (18 females, 11 males) walked at their preferred walking speed on three different surfaces (asphalt, cobbles, grass) in single-task and in two dual-task conditions (manual task—carrying a cup filled with water, cognitive task—subtracting the number 7). A triaxial inertial sensor attached to the lower trunk was used to record trunk acceleration during gait. From 15 strides, sample entropy (SampEn) as an indicator of gait complexity and root mean square (RMS) as an indicator of gait variability were computed. The findings demonstrate that in an outdoor environment, the surfaces significantly impacted only gait variability, not complexity, and that the tasks affected both gait variability and complexity in young healthy adults.

DEFORESTATION DETECTION IN THE AMAZON RAINFOREST WITH SPATIAL AND CHANNEL ATTENTION MECHANISMS

Deforestation in the Amazon rainforest is an alarming problem of global interest. Environmental impacts of this process are countless, but probably the most significant concerns regard the increase in CO2 emissions and global temperature rise. Currently, the assessment of deforested areas in the Amazon region is a manual task, where people analyse multiple satellite images to quantify the deforestation. We propose a method for automatic deforestation detection based on Deep Learning Neural Networks with dual-attention mechanisms. We employed a siamese architecture to detect deforestation changes between optical images in 2018 and 2019. Experiments were performed to evaluate the relevance and sensitivity of hyperparameter tuning of the loss function and the effects of dual-attention mechanisms (spatial and channel) in predicting deforestation. Experimental results suggest that a proper tuning of the loss function might bring benefits in terms of generalisation. We also show that the spatial attention mechanism is more relevant for deforestation detection than the channel attention mechanism. When both mechanisms are combined, the greatest improvements are found, and we reported an increase of 1.06% in the mean average precision over a baseline.

Automated scoring of pre-REM sleep in mice with deep learning

Reliable automation of the labor-intensive manual task of scoring animal sleep can facilitate the analysis of long-term sleep studies. In recent years, deep-learning-based systems, which learn optimal features from the data, increased scoring accuracies for the classical sleep stages of Wake, REM, and Non-REM. Meanwhile, it has been recognized that the statistics of transitional stages such as pre-REM, found between Non-REM and REM, may hold additional insight into the physiology of sleep and are now under vivid investigation. We propose a classification system based on a simple neural network architecture that scores the classical stages as well as pre-REM sleep in mice. When restricted to the classical stages, the optimized network showed state-of-the-art classification performance with an out-of-sample F1 score of 0.95 in male C57BL/6J mice. When unrestricted, the network showed lower F1 scores on pre-REM (0.5) compared to the classical stages. The result is comparable to previous attempts to score transitional stages in other species such as transition sleep in rats or N1 sleep in humans. Nevertheless, we observed that the sequence of predictions including pre-REM typically transitioned from Non-REM to REM reflecting sleep dynamics observed by human scorers. Our findings provide further evidence for the difficulty of scoring transitional sleep stages, likely because such stages of sleep are under-represented in typical data sets or show large inter-scorer variability. We further provide our source code and an online platform to run predictions with our trained network.

Quantification of Osteoclasts in Culture, Powered by Machine Learning

In vitro osteoclastogenesis is a central assay in bone biology to study the effect of genetic and pharmacologic cues on the differentiation of bone resorbing osteoclasts. To date, identification of TRAP+ multinucleated cells and measurements of osteoclast number and surface rely on a manual tracing requiring specially trained lab personnel. This task is tedious, time-consuming, and prone to operator bias. Here, we propose to replace this laborious manual task with a completely automatic process using algorithms developed for computer vision. To this end, we manually annotated full cultures by contouring each cell, and trained a machine learning algorithm to detect and classify cells into preosteoclast (TRAP+ cells with 1–2 nuclei), osteoclast type I (cells with more than 3 nuclei and less than 15 nuclei), and osteoclast type II (cells with more than 15 nuclei). The training usually requires thousands of annotated samples and we developed an approach to minimize this requirement. Our novel strategy was to train the algorithm by working at “patch-level” instead of on the full culture, thus amplifying by &gt;20-fold the number of patches to train on. To assess the accuracy of our algorithm, we asked whether our model measures osteoclast number and area at least as well as any two trained human annotators. The results indicated that for osteoclast type I cells, our new model achieves a Pearson correlation (r) of 0.916 to 0.951 with human annotators in the estimation of osteoclast number, and 0.773 to 0.879 for estimating the osteoclast area. Because the correlation between 3 different trained annotators ranged between 0.948 and 0.958 for the cell count and between 0.915 and 0.936 for the area, we can conclude that our trained model is in good agreement with trained lab personnel, with a correlation that is similar to inter-annotator correlation. Automation of osteoclast culture quantification is a useful labor-saving and unbiased technique, and we suggest that a similar machine-learning approach may prove beneficial for other morphometrical analyses.

Development of a Semiautomatic Parametric Method for Creation of an I-BIM Model of a Tunnel for Use in FEM Software

This paper focuses on the design phase of I-BIM tunnelling projects using Sequential Excavation Method (SEM), in Europe, commonly referred to as the New Austrian Tunnelling Method (NATM), and addresses the problem of coupling geotechnical conditions and tunnelling Building Information Model (BIM) for the preparation of the computational model suitable for the finite element analysis method (FEM). The review of the literature led to the conclusion that an automatic merging of the tunnel model and ground model for use in the FEM software is currently not reliable due to the number of differences between various types of models as they serve contrasting needs. Consequently, modelling becomes a manual task, which is very time-consuming and error prone. In this paper, we present the development of a framework for the semiautomatic transformation of the various tunnelling models and respective ground models into the model suitable for further analysis. We conclude that the import and translation of the geometry into the FEM software are most successful and accurate when the initial I- BIM (tunnel) model is prepared at a level of detail appropriate for a computational model. The result is the I-BIM model, fit for use in the FEM software which speeds up the modelling process and reduces errors. We have shown that it is possible to prepare the geometry of a tunnel in the BIM software, transfer it, and use it in the software for geotechnical analysis. This makes the preparation of the tunnel geometry for FEM analysis much easier and faster. Due to the fast preparation of the geometry of the new model, the approach presented in this research is useful in practice. The applicability of the framework and the framework workflow are both presented through a practical case study.