The exponential growth of user-generated content has increased the need for efficient video summarization schemes. However, most approaches underestimate the power of aural features, while they are designed to work mainly on commercial/professional videos. In this work, we present an approach that uses both aural and visual features in order to create video summaries from user-generated videos. Our approach produces dynamic video summaries, that is, comprising the most “important” parts of the original video, which are arranged so as to preserve their temporal order. We use supervised knowledge from both the aforementioned modalities and train a binary classifier, which learns to recognize the important parts of videos. Moreover, we present a novel user-generated dataset which contains videos from several categories. Every 1 s part of each video from our dataset has been annotated by more than three annotators as being important or not. We evaluate our approach using several classification strategies based on audio, video and fused features. Our experimental results illustrate the potential of our approach.

Egocentric activity recognition is a prominent computer vision task that is based on the use of wearable cameras. Since egocentric videos are captured through the perspective of the person wearing the camera, her/his body motions severely complicate the video content, imposing several challenges. In this work we propose a novel approach for domain-generalized egocentric human activity recognition. Typical approaches use a large amount of training data, aiming to cover all possible variants of each action. Moreover, several recent approaches have attempted to handle discrepancies between domains with a variety of costly and mostly unsupervised domain adaptation methods. In our approach we show that through simple manipulation of available source domain data and with minor involvement from the target domain, we are able to produce robust models, able to adequately predict human activity in egocentric video sequences. To this end, we introduce a novel three-stream deep neural network architecture combining elements of vision transformers and residual neural networks which are trained using multi-modal data. We evaluate the proposed approach using a challenging, egocentric video dataset and demonstrate its superiority over recent, state-of-the-art research works.

Monitoring driving behaviour is important in controlling driving risk, fuel consumption, and CO₂ emissions. Recent advances in machine learning, which include several variants of convolutional neural networks (CNNs), and recurrent neural networks (RNNs), such as long short-term memory (LSTM) and gated recurrent unit (GRU) networks, could be valuable for the development of objective and efficient computational tools in this direction. The main idea in this work is to complement data-driven classification of driving behaviour with rules derived from domain knowledge. In this light, we present a hybrid representation approach, which employs NN-based time-series encoding and rule-guided event detection. Histograms derived from the output of these two components are concatenated, normalized, and used to train a standard support vector machine (SVM). For the NN-based component, CNN-based, LSTM-based, and GRU-based variants are investigated. The CNN-based variant uses image-like representations of sensor measurements, whereas the RNN-based variants (LSTM and GRU) directly process sensor measurements in the form of time-series. Experimental evaluation on three datasets leads to the conclusion that the proposed approach outperforms a state-of-the-art camera-based approaches in distinguishing between normal and aggressive driving behaviour without using data derived from a camera. Moreover, it is demonstrated that both NN-guided time-series encoding and rule-guided event detection contribute to overall classification accuracy.

During the last few years, several technological advances have led to an increase in the creation and consumption of audiovisual multimedia content. Users are overexposed to videos via several social media or video sharing websites and mobile phone applications. For efficient browsing, searching, and navigation across several multimedia collections and repositories, e.g., for finding videos that are relevant to a particular topic or interest, this ever-increasing content should be efficiently described by informative yet concise content representations. A common solution to this problem is the construction of a brief summary of a video, which could be presented to the user, instead of the full video, so that she/he could then decide whether to watch or ignore the whole video. Such summaries are ideally more expressive than other alternatives, such as brief textual descriptions or keywords. In this work, the video summarization problem is approached as a supervised classification task, which relies on feature fusion of audio and visual data. Specifically, the goal of this work is to generate dynamic video summaries, i.e., compositions of parts of the original video, which include its most essential video segments, while preserving the original temporal sequence. This work relies on annotated datasets on a per-frame basis, wherein parts of videos are annotated as being “informative” or “noninformative”, with the latter being excluded from the produced summary. The novelties of the proposed approach are, (a) prior to classification, a transfer learning strategy to use deep features from pretrained models is employed. These models have been used as input to the classifiers, making them more intuitive and robust to objectiveness, and (b) the training dataset was augmented by using other publicly available datasets. The proposed approach is evaluated using three datasets of user-generated videos, and it is demonstrated that deep features and data augmentation are able to improve the accuracy of video summaries based on human annotations. Moreover, it is domain independent, could be used on any video, and could be extended to rely on richer feature representations or include other data modalities.

The presence of occlusion in human activity recognition (HAR) tasks hinders the performance of recognition algorithms, as it is responsible for the loss of crucial motion data. Although it is intuitive that it may occur in almost any real-life environment, it is often underestimated in most research works, which tend to rely on datasets that have been collected under ideal conditions, i.e., without any occlusion. In this work, we present an approach that aimed to deal with occlusion in an HAR task. We relied on previous work on HAR and artificially created occluded data samples, assuming that occlusion may prevent the recognition of one or two body parts. The HAR approach we used is based on a Convolutional Neural Network (CNN) that has been trained using 2D representations of 3D skeletal motion. We considered cases in which the network was trained with and without occluded samples and evaluated our approach in single-view, cross-view, and cross-subject cases and using two large scale human motion datasets. Our experimental results indicate that the proposed training strategy is able to provide a significant boost of performance in the presence of occlusion.

Recognizing human activities from motion data is a complex task in computer vision, involving the recognition of human behaviors from sequences of 3D motion data. These activities encompass successive body part movements, interactions with objects, or group dynamics. Camera-based recognition methods are cost-effective and perform well under controlled conditions but face challenges in real-world scenarios due to factors such as viewpoint changes, illumination variations, and occlusion. The latter is the most significant challenge in real-world recognition; partial occlusion impacts recognition accuracy to varying degrees depending on the activity and the occluded body parts while complete occlusion can render activity recognition impossible. In this paper, we propose a novel approach for human activity recognition in the presence of partial occlusion, which may be applied in cases wherein up to two body parts are occluded. The proposed approach works under the assumptions that (a) human motion is modeled using a set of 3D skeletal joints, and (b) the same body parts remain occluded throughout the whole activity. Contrary to previous research, in this work, we address this problem using a Generative Adversarial Network (GAN). Specifically, we train a Convolutional Recurrent Neural Network (CRNN), whose goal is to serve as the generator of the GAN. Its aim is to complete the missing parts of the skeleton due to occlusion. Specifically, the input to this CRNN consists of raw 3D skeleton joint positions, upon the removal of joints corresponding to occluded parts. The output of the CRNN is a reconstructed skeleton. For the discriminator of the GAN, we use a simple long short-term memory (LSTM) network. We evaluate the proposed approach using publicly available datasets in a series of occlusion scenarios. We demonstrate that in all scenarios, the occlusion of certain body parts causes a significant decline in performance, although in some cases, the reconstruction process leads to almost perfect recognition. Nonetheless, in almost every circumstance, the herein proposed approach exhibits superior performance compared to previous works, which varies between $2.2\%$ and $37.5\%$, depending on the dataset used and the occlusion case.