News & publications

An important function of the construction of the Brain-Computer Interface (BCI) device is the development of a model that is able to recognize emotions from electroencephalogram (EEG) signals. Research in this area is very challenging because the EEG signal is non-stationary, non-linear, and contains a lot of noise due to artifacts caused by muscle activity and poor electrode contact. EEG signals are recorded with non-invasive wearable devices using a large number of electrodes, which increase the dimensionality and, thereby, also the computational complexity of EEG data. It also reduces the level of comfort of the subjects. This paper implements our holographic features, investigates electrode selection, and uses the most relevant channels to maximize model accuracy. The ReliefF and Neighborhood Component Analysis (NCA) methods were used to select the optimal electrodes. Verification was performed on four publicly available datasets. Our holographic feature maps were constructed using computer-generated holography (CGH) based on the values of signal characteristics displayed in space. The resulting 2D maps are the input to the Convolutional Neural Network (CNN), which serves as a feature extraction method. This ethodology uses a reduced set of electrodes, which are different between men and women, and obtains state-of-the-art results in a three-dimensional motional space. The experimental results show that the channel selection methods improve emotion recognition rates significantly with an accuracy of 90.76% for valence, 92.92% for arousal, and 92.97% for dominance.

The paper presents the usage of artificial neural networks (NNs) in bicycle route planning. This research aimed to check the possibility of NNs to transfer human expertise in bicycle route design by training the NN on an already established set of bicycle routes and then using the trained NN to design the routes on the novel area. We created two NNs capable of choosing the best route among the given road network by training them on two different areas. The bicycle routes produced by NNs were the same at best and had 75% overlap at the worst compared to those produced by human experts. Furthermore, the mean square error for all of our NN models varied from 0.015 and 0.081. We compared this new approach to the traditional multicriteria GIS (geographic information system) analysis (MA) that requires the human expert to define the bicycle route selection criteria. The benefit of using NN over the MA was that the NN directly transfers the human expertise to a model. In contrast, the MA needs the expert to select multiple criteria and adjust their weights carefully.

Acoustic properties affect the way one experiences sound in a room; sound listener or microphone will receive the sound in a form that might be more or less understandable, influenced by the room’s acoustic properties. Assessment of a room’s acoustic properties can be made on-site and by using simulation software. This paper used both methods for acoustic properties assessment of rooms with various dimensions and purposes. The aim was to compare the results obtained by both methods and advise on how to improve the room’s acoustic properties. Acoustic properties are assessed using: T20, C50, and C80. The difference in the observed room’s acoustic is presented. The results showed a significant agreement between acoustic properties measured on-site and by simulation.

People detection in image and video is challenging problem that has great importance in different applications such as surveillance systems, autonomous driving systems, sports video analysis etc. Player detection task, as a subproblem of people detection, is one of the fundamental steps in football video analysis. In this paper we propose a method for player detection based on a fully convolutional neural network. Novelty in our approach is usage of U-Net architecture for generation of player probability map. U-Net consists of contracting path that has typical convolutional neural network architecture and extracting path where upsampled feature maps are combined with features from contracting path to obtain more precise segmentation. Next step is thresholding of player probability map followed by connected component analysis that gives player bounding boxes. Experimental results show promising performance on football field images including distant views, motion blur, complex background etc.

Emotion recognition using electroencephalogram (EEG) signals is getting more and more attention in recent years. Since the EEG signals are noisy, non-linear and have non-stationary properties, it is a challenging task to develop an intelligent framework that can provide high accuracy for emotion recognition. In this paper, we propose a new model for emotion recognition that will be based on the creation of feature maps based on the topographic (TOPO-FM) and holographic (HOLO-FM) representation of EEG signal characteristics. Deep learning has been utilized as a feature extractor method on feature maps, and afterward extracted features are fused together for the classification process to recognize different kinds of emotions. The experiments are conducted on the four publicly available emotion datasets: DEAP, SEED, DREAMER, and AMIGOS. We demonstrated the effectiveness of our approaches in comparison with studies where authors used EEG signals that classify human emotions in the two-dimensional space. Experimental results show that the proposed methods can improve the emotion recognition rate on the different size datasets.