Crowd video monitoring and analysis is a hot topic in computer vision and public management. The pre-evaluation of crowd safety is beneficial to the prediction of crowd status to avoid the occurrence of catastrophic events. This paper proposes a method to evaluate crowd safety based on fuzzy inference. Pedestrian’s number and distribution uniformity are considered in a fuzzy inference system as two kinds of attributes of a crowd. Firstly, the pedestrian’s number is estimated by the number of foreground pixels. Then, the distribution uniformity of a crowd is calculated using distribution entropy by dividing the monitoring scene into several small areas. Furthermore, through the fuzzy operation, the fuzzy system is constructed by using two input variables (pedestrian’s number and distribution entropy) and one output variable (crowd safety status). Finally, inference rules between the crowd safety state and the pedestrian’s number and distribution uniformity are constructed to obtain the pre-evaluation of the safety state of the crowd. Three video sequences extracted from different scenes are used in the experiment. Experimental results show that the proposed method can be used to evaluate the safety status of the crowd in a monitoring scene.

Due to the rapid development of human–computer interaction, affective computing has attracted more and more attention in recent years. In emotion recognition, Electroencephalogram (EEG) signals are easier to be recorded than other physiological experiments and are not easily camouflaged. Because of the high dimensional nature of EEG data and the diversity of human emotions, it is difficult to extract effective EEG features and recognize the emotion patterns. This paper proposes a multi-feature deep forest (MFDF) model to identify human emotions. The EEG signals are firstly divided into several EEG frequency bands and then extract the power spectral density (PSD) and differential entropy (DE) from each frequency band and the original signal as features. A five-class emotion model is used to mark five emotions, including neutral, angry, sad, happy, and pleasant. With either original features or dimension reduced features as input, the deep forest is constructed to classify the five emotions. These experiments are conducted on a public dataset for emotion analysis using physiological signals (DEAP). The experimental results are compared with traditional classifiers, including K Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Machine (SVM). The MFDF achieves the average recognition accuracy of 71.05%, which is 3.40%, 8.54%, and 19.53% higher than RF, KNN, and SVM, respectively. Besides, the accuracies with the input of features after dimension reduction and raw EEG signal are only 51.30 and 26.71%, respectively. The result of this study shows that the method can effectively contribute to EEG-based emotion classification tasks.

Facial expressions are an important part of human emotional signals and their recognition has become an important topic of research in the field of pattern recognition. Deep learning based methods have achieved great success in the recognition of facial expressions. However, with the evolution of convolution neural networks and the increased network depth, these methods suffer from problems such as degraded network performance and loss of feature information. To address these problems, a novel facial expression recognition algorithm based on an improved residual neural network is proposed. First, a residual neural network is designed to extract deep features while retaining the shallow ones. This can effectively prevent the degradation of network performance. Moreover, when the gradient of the Rectified Linear Units activation function used in the residual module is 0, it will inactivate the neurons and cause a loss of feature information. To address this, the Mish activation function is used instead. The slight allowance for negative values in Mish improves the gradient flow. Next, an inception module is introduced to obtain richer feature information under the same receptive field. Finally, by conducting verification experiments on the public datasets CK+ and KDEF, the authors manage to solve the problems of degraded network performance and insufficient information from extracted features, achieving recognition accuracy rates of 96.37% and 93.38% on the two datasets, respectively.

The change of crowd energy is a fundamental measurement for describing a crowd behavior. In this paper, we present a crowd abnormal detection method based on the change of energy-level distribution. The method can not only reduce the camera perspective effect, but also detect crowd abnormal behavior in time. Pixels in the image are treated as particles, and the optical flow method is adopted to extract the velocities of particles. The qualities of different particles are distributed as different value according to the distance between the particle and the camera to reduce the camera perspective effect. Then a crowd motion segmentation method based on flow field texture representation is utilized to extract the motion foreground, and a linear interpolation calculation is applied to pedestrian’s foreground area to determine their distance to the camera. This contributes to the calculation of the particle qualities in different locations. Finally, the crowd behavior is analyzed according to the change of the consistency, entropy and contrast of the three descriptors for co-occurrence matrix. By calculating a threshold, the timestamp when the crowd abnormal happens is determined. In this paper, multiple sets of videos from three different scenes in UMN dataset are employed in the experiment. The results show that the proposed method is effective in characterizing anomalies in videos.

Recent single image super-resolution methods based on various complex deep neural networks have achieved remarkable success. However, these methods require a large amount of computational overhead while improving performance, and thus are difficult to apply to mobile devices in real-world scenarios. In this letter, we design an efficient asymmetric convolutional distillation block (ACDB). Especially in this block, introducing an asymmetric convolution block (ACB) and reusing shallow distillation features can effectively improve the performance of the model and reduce the model complexity. Our model achieves efficient performance while maintaining low complexity.>