Surface electromyography (sEMG) is an important measurement technique used in biomechanical, rehabilitation and sport environments. In this article the design, development and testing of a low-cost wearable sEMG system are described. The hardware architecture consists of a two-cascade small-sized bioamplifier with a total gain of 2,000 and band-pass of 3 to 500 Hz. The sampling frequency of the system is 1,000 Hz. Since real measured EMG signals are usually corrupted by various types of noises (motion artifacts, white noise and electromagnetic noise present at 50 Hz and higher harmonics), we have tested several denoising techniques, both on artificial and measured EMG signals. Results showed that a wavelet—based technique implementing Daubechies5 wavelet and soft sqtwolog thresholding is the most appropriate for EMG signals denoising. To test the system performance, EMG activities of six dominant muscles of ten healthy subjects during gait were measured (gluteus maximus, biceps femoris, sartorius, rectus femoris, tibialis anterior and medial gastrocnemius). The obtained EMG envelopes presented against the duration of gait cycle were compared favourably with the EMG data available in the literature, suggesting that the proposed system is suitable for a wide range of applications in biomechanics.

In this paper an approach to classify hand shapes into different classes according to the similarity measures between features is proposed. We show how to use an Exploratory Data Analysis to extract novel, single feature of hand from images. Based on the obtained curve-like shape of the feature, hands are classified into one of 21 possible classes of Croatian sign language using Dynamic Time Warping and Longest Common Subsequence as similarity measures. Performance of the system was evaluated with 1260 images. Results show that high classification accuracy can be obtained from a single feature recognition and a small number of training sample.

The main goal of any classification or regression task is to obtain a model that will generalize well on new, previously unseen data. Due to the recent rise of deep learning and many state-of-the-art results obtained with deep models, deep learning architectures have become one of the most used model architectures nowadays. To generalize well, a deep model needs to learn the training data well without overfitting. The latter implies a correlation of deep model optimization and regularization with generalization performance. In this work, we explore the effect of the used optimization algorithm and regularization techniques on the final generalization performance of the model with convolutional neural network (CNN) architecture widely used in the field of computer vision. We give a detailed overview of optimization and regularization techniques with a comparative analysis of their performance with three CNNs on the CIFAR-10 and Fashion-MNIST image datasets.

The authors propose a novel image segmentation technique based on the non-parametric clustering procedure in the discretised colour space. The discrete probability density function is estimated in two steps. Multidimensional colour histogram is created, which is afterwards used to acquire final density estimate using the variable kernel density estimation technique. Segmentation is obtained by mapping revealed range domain clusters to the spatial image domain. The proposed method is highly efficient, running in time linear to the number of the image pixels with low constant factors. The output of the algorithm can be accommodated for a particular application to simplify the integration with other image processing techniques. Quantitative evaluation on a standard test dataset proves that the quality of the segmentations provided by the proposed method is comparable to the quality of the segmentations generated by other widely adopted low-level segmentation techniques, while running times are several times faster.

Laser pointers are one of the most widely used interactive and pointing devices in different human-computer interaction systems. Existing approaches to vision-based laser spot tracking are designed for controlled indoor environments with the main assumption that the laser spot is very bright, if not the brightest, spot in images. In this work, we are interested in developing a method for an outdoor, open-space environment, which could be implemented on embedded devices with limited computational resources. Under these circumstances, none of the assumptions of existing methods for laser spot tracking can be applied, yet a novel and fast method with robust performance is required. Throughout the paper, we will propose and evaluate an efficient method based on modified circular Hough transform and Lucas–Kanade motion analysis. Encouraging results on a representative dataset demonstrate the potential of our method in an uncontrolled outdoor environment, while achieving maximal accuracy indoors. Our dataset and ground truth data are made publicly available for further development.

An integrative approach to maritime accident risk factor assessment in accordance with formal safety assessment is proposed, which exploits the multifaceted capabilities of Bayesian networks (BNs) by consolidation of modelling, verification, and validation. The methodology for probabilistic modelling with BNs is well known and its application to risk assessment is based on the model verified though sensitivity analysis only, while validation of the model is often omitted due to a lack of established evaluation measures applicable to scarce real-world data. For this reason, in this work, the modified Lyapunov divergence measure is proposed as a novel quantitative assessor that can be efficiently exploited on an individual accident scenario for contributing causal factor identification, and thus can serve as the measure for validation of the developed expert elicited BN. The proposed framework and its approach are showcased for maritime grounding of small passenger ships in the Adriatic, with the complete grounding model disclosed, quantitative validation performed, and its utilization for causal factor identification and risk factor ranking presented. The data from two real-world grounding cases demonstrate the explanatory capabilities of the developed approach.