Search

Acute alcohol intoxication impairs cognitive and psychomotor abilities leading to various public health hazards such as road traffic accidents and alcohol-related violence. Intoxicated individuals are usually identified by measuring their blood alcohol concentration (BAC) using breathalysers that are expensive and labour-intensive. In this paper, we developed the Audio-based Deep Learning Algorithm to Identify Alcohol Inebriation (ADLAIA) that can instantly predict an individual's intoxication status based on a 12-second recording of their speech.ADLAIA was trained on a publicly available German Alcohol Language Corpus that comprises a total of 12,360 audio clips of inebriated and sober speakers (total of 162, aged 21-64, 47.7% female). ADLAIA's performance was determined by computing the unweighted average recall (UAR) and accuracy of inebriation prediction.ADLAIA was able to identify inebriated speakers-with BAC of 0.05% or higher-with an UAR of 68.09% and accuracy of 67.67%. ADLAIA had a higher performance (UAR of 75.7%) in identifying intoxicated speakers (BAC0.12%).Being able to identify intoxicated individuals solely based on their speech, ADLAIA could be integrated in mobile applications and used in environments (such as bars, sports stadiums) to get instantaneous results about inebriation status of individuals.

Purpose: Increasing access to marker-less technology has enabled practitioners to obtain kinematic data more quickly. However, the validation of many of these methods is lacking. Therefore, the validity of pre-trained neural networks was explored in this study compared to reflective marker tracking from sagittal plane cycling motion. Methods: Twenty-six cyclists were assessed during stationary cycling at self-selected cadence and moderate intensity exercise. Standard video from their sagittal plane was obtained to extract joint kinematics. Hip, knee, and ankle angles were calculated from marker digitisation and from two deep learning-based approaches (TransPose and MediaPipe). Results: Typical errors ranged between 1-10°for TransPose and 3-9°for MediaPipe. Correlations between joint angles calculated from TransPose and marker digitalization were stronger (0.47-0.98) than those from MediaPipe (0.25-0.96). Conclusion: TransPose seemed to perform better than MediaPipe but both methods presented poor performance when tracking the foot and ankle. This seems to be associated with the low frame rate and image resolution when using standard video mode.

Advances in deep neural networks have led to significant improvement of object detection accuracy. However, object detection in crowded scenarios is a challenging task for neural networks since extremely overlapped objects provide fewer visible cues for a model to learn from. Further complicating the detection of overlapping objects is the fact that most object detectors produce multiple redundant detections for single objects, which are indistinguishable from detections of separate overlapped objects. Most existing works use some variant of non-maximum suppression to prune duplicate candidate bounding boxes based on their confidence scores and the amount of overlap between predicted bounding boxes. These methods are unaware of how much overlap there actually is between the objects in the image, and are therefore inclined to merge detections for highly overlapped objects. In this paper, we propose an overlap aware box selection solution that uses a predicted overlap map to help it decide which highly overlapping bounding boxes are associated with actual overlapping objects and should not be pruned. We show our solution outperforms the state-of-the-art set-NMS bounding box selection algorithm for both the crowdHuman dataset and a sports dataset.

We introduce a novel deep learning based group activity recognition approach called the Pose Only Group Activity Recognition System (POGARS), designed to use only tracked poses of people to predict the performed group activity. In contrast to existing approaches for group activity recognition, POGARS uses 1D CNNs to learn spatiotemporal dynamics of individuals involved in a group activity and forgo learning features from pixel data. The proposed model uses a spatial and temporal attention mechanism to infer person-wise importance and multi-task learning for simultaneously performing group and individual action classification. Experimental results confirm that POGARS achieves highly competitive results compared to state-of-the-art methods on a widely used public volleyball dataset despite only using tracked pose as input. Further our experiments show by using pose only as input, POGARS has better generalization capabilities compared to methods that use RGB as input., 12 pages, 7 figures

It is very important for swimming coaches to analyse a swimmer’s performance at the end of each race, since the analysis can then be used to change strategies for the next round. Coaches rely heavily on statistics, such as stroke length and instantaneous velocity, when analysing performance. These statistics are usually derived from time-consuming manual video annotations. To automatically obtain the required statistics from swimming videos, we need to solve the following four challenging computer vision tasks: swimmer head detection; tracking; stroke detection; and camera calibration. We collectively solve these problems using a two-phased deep learning approach, we call Deep Detector for Actions and Swimmer Heads (DeepDASH). DeepDASH achieves a 20.8% higher F1 score for swimmer head detection and operates 6 times faster than the popular Faster R-CNN object detector. We also propose a hierarchical tracking algorithm based on the existing SORT algorithm which we call HISORT. HISORT produces significantly longer tracks than SORT by preserving swimmer identities for longer periods of time. Finally, DeepDASH achieves an overall F1 score of 97.5% for stroke detection across all four swimming stroke styles.

Medical image classification is often challenging for two reasons: a lack of labelled examples due to expensive and time-consuming annotation protocols, and imbalanced class labels due to the relative scarcity of disease-positive individuals in the wider population. Semi-supervised learning (SSL) methods exist for dealing with a lack of labels, but they generally do not address the problem of class imbalance. In this study we propose Adaptive Blended Consistency Loss (ABCL), a drop-in replacement for consistency loss in perturbation-based SSL methods. ABCL counteracts data skew by adaptively mixing the target class distribution of the consistency loss in accordance with class frequency. Our experiments with ABCL reveal improvements to unweighted average recall on two different imbalanced medical image classification datasets when compared with existing consistency losses that are not designed to counteract class imbalance., This paper has 28 pages, 7 figures

Commensal rats and mice are a major pest. Commercial-scale control measures require frequent manual checking of rodenticide levels in bait stations and searching for evidence of activity. In this paper, we propose a low power remote monitoring system, which has a projected battery life of over 12 months, which wirelessly transmits bait and battery data. The ability of our system to detect and communicate rodenticide depletion was tested under laboratory and field conditions, with initial results showing the potential of our approach.

Sophisticated trajectory prediction models that effectively mimic team dynamics have many potential uses for sports coaches, broadcasters and spectators. However, through experiments on soccer data we found that it can be surprisingly challenging to train a deep learning model for player trajectory prediction which outperforms linear extrapolation on average distance between predicted and true future trajectories. We propose and test a novel method for improving training by predicting a sparse trajectory and interpolating using constant acceleration, which improves performance for several models. This interpolation can also be used on models that aren't trained with sparse outputs, and we find that this consistently improves performance for all tested models. Additionally, we find that the accuracy of predicted trajectories for a subset of players can be improved by conditioning on the full trajectories of the other players, and that this is further improved when combined with sparse predictions. We also propose a novel architecture using graph networks and multi-head attention (GraN-MA) which achieves better performance than other tested state-of-the-art models on our dataset and is trivially adapted for both sparse trajectories and full-trajectory conditioned trajectory prediction., Comment: 10 pages (not including references), 7 figures. Published in Neural Computing and Applications on 20 March 2021

Recent advances in deep learning approaches to computer vision problems have led to renewed interest in the task of predicting 3D human joint locations from raw image data, with application areas including sports analysis, human-computer interaction, and physical rehabilitation. Although supervised learning of deep neural networks has proven to be effective for pose estimation, it requires a wealth of varied data to generalise well to previously unseen examples. Consequently, progress in 3D human pose estimation has been slowed by the fact that 3D keypoint annotations are notoriously difficult to obtain, traditionally requiring a large array of cameras and/or the use of wearable markers/sensors. In this paper we describe a methodology for obtaining 3D human pose annotations using only three video cameras and without any wearables. We apply this methodology to construct ASPset-510 (Australian Sports Pose Dataset), a large collection of natural sports-related video with 3D pose annotations. Using ASPset-510 as an additional source of training examples, we found that we could improve pose model generalisation on the established MPI-INF-3DHP benchmark. We make ASPset-510 publicly available, and provide strong baseline results for future work to compare against.

Multiple object tracking is an important but challenging computer vision problem. The complex motion of objects makes tracking difficult during long periods of object occlusion, and as a result occlusions frequently cause fragmented tracks with gaps. Previous works use linear interpolation to fill in such gaps, a technique which is only able to model simple motion. As a result, tracked bounding box locations can be quite poor in these situations. In this paper, we propose a 1D CNN based solution to filling gaps which models complex motion in a data-driven way. Our proposed solution uses only bounding box coordinates as input, and as such does not incur the computational cost of processing image features directly. We show that our model significantly outperforms linear interpolation on dynamic sports datasets in terms of mean intersection over union between predicted and ground truth bounding boxes.

Automatically determining three-dimensional human pose from monocular RGB image data is a challenging problem. The two-dimensional nature of the input results in intrinsic ambiguities which make inferring depth particularly difficult. Recently, researchers have demonstrated that the flexible statistical modelling capabilities of deep neural networks are sufficient to make such inferences with reasonable accuracy. However, many of these models use coordinate output techniques which are memory-intensive, not differentiable, and/or do not spatially generalise well. We propose improvements to 3D coordinate prediction which avoid the aforementioned undesirable traits by predicting 2D marginal heatmaps under an augmented soft-argmax scheme. Our resulting model, MargiPose, produces visually coherent heatmaps whilst maintaining differentiability. We are also able to achieve state-of-the-art accuracy on publicly available 3D human pose estimation data., Accepted in WACV 2019

The need to store vast amounts of trajectory data becomes more problematic as GPS-based tracking devices become increasingly prevalent. There are two commonly used approaches for compressing trajectory data. The first is the line generalisation approach which aims to fit the trajectory using a series of line segments. The second is to store the initial data point and then store the remaining data points as a sequence of successive deltas. The line generalisation approach is only effective when given a large error margin, and existing delta compression algorithms do not permit lossy compression. Consequently there is an uncovered gap in which users expect a good compression ratio by giving away only a small error margin. This paper fills this gap by extending the delta compression approach to allow users to trade a small maximum error margin for large improvements to the compression ratio. In addition, alternative techniques are extensively studied for the following two key components of any delta-based approach: predicting the value of the next data point and encoding leading zeros. We propose a new trajectory compression system called Trajic based on the results of the study. Experimental results show that Trajic produces 1.5 times smaller compressed data than a straight-forward delta compression algorithm for lossless compression and produces 9.4 times smaller compressed data than a state-of-the-art line generalisation algorithm when using a small maximum error bound of 1 meter.

PURPOSE : Lung cancer has the highest death rate among all cancers in the USA. In this work we focus on improving the ability of computer-aided diagnosis (CAD) systems to predict the malignancy of nodules from cropped CT images of lung nodules.We evaluate the effectiveness of very deep convolutional neural networks at the task of expert-level lung nodule malignancy classification. Using the state-of-the-art ResNet architecture as our basis, we explore the effect of curriculum learning, transfer learning, and varying network depth on the accuracy of malignancy classification.Due to a lack of public datasets with standardized problem definitions and train/test splits, studies in this area tend to not compare directly against other existing work. This makes it hard to know the relative improvement in the new solution. In contrast, we directly compare our system against two state-of-the-art deep learning systems for nodule classification on the LIDC/IDRI dataset using the same experimental setup and data set. The results show that our system achieves the highest performance in terms of all metrics measured including sensitivity, specificity, precision, AUROC, and accuracy.The proposed method of combining deep residual learning, curriculum learning, and transfer learning translates to high nodule classification accuracy. This reveals a promising new direction for effective pulmonary nodule CAD systems that mirrors the success of recent deep learning advances in other image-based application domains.