Search

This paper presents a novel method for global anomaly detection in crowded scenes. The optical flow of frames is used to extract the foreground of areas with people motions in the crowd in the form of layers. The optical flow between two frames generates one layer. The proposed method applies the metaheuristic of artificial bacteria colony as a robust algorithm to optimize the extracted layers. Artificial bacteria cover all regions of interest that have high movement between frames. The artificial bacteria colony adapts quickly to the most varied scenarios. Moreover, the algorithm has low sensibility to noise and to sudden changes in video lighting as captured by optical flow. The bacteria population of the colonies, its food storage and the colony’s centroid position regarding each optical flow layer, are used as input to train a Kohonen’s neural network. Once trained the network is able to detect specific events based on behavior patterns similarity, as produced by the bacteria colony during such events. Experiments are conducted on available public dataset. The achieved results show that the proposed method captures the dynamics of the crowd behavior successfully, revealing that the proposed scheme outperforms the available state-of-the-art algorithms for global anomaly detection. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, we propose a massively parallel architecture for hardware implementation of genetic algorithms. This design is quite innovative as it provides a viable solution to the fitness computation problem, which depends heavily on the problem-specific knowledge. The proposed architecture is completely independent of such specifics. It implements the fitness computation using a neural network. The hardware implementation of the used neural network is stochastic and thus minimise the required hardware area without much increase in response time. Last but not least, we demonstrate the characteristics of the proposed hardware and compare it to existing ones. [Copyright &y& Elsevier]

The automatic speech recognition based on detection of phonemes provides advantages for online recognition of a speech represented by a sound signal. The development of a system for automatic speech recognition is multidisciplinary. It covers several areas of research, such as linguistics, signal processing and computational intelligence. In this work, the process starts with a speech signal pre-processing to extract the main features of the speech signal at a given instant of time. Inspired by the "divide and conquer" principle, we bridge the complexity gap of automatic speech recognition by devising models based on an ensemble of neural network experts, allowing to divide the huge decision space regarding speech recognition so that each expert takes care only of a delimited area of this decision space. This novel application of this strategy improves the precision, sensitivity and accuracy of the recognition process. Each included expert decides regarding each one of the pre-processed input samples. The decision set thus obtained is weighted. So, the expert with the highest weight for the output will determine the sample final classification. After that, a dynamic post-processing step, implemented as a recurrent neural network, is executed. It aims at mitigating the oscillatory effect that occurs during the recognition of classes with similar characteristics. In this work, two ensembles are investigated. The first is based on the clustering of similar phonetics classes while the second takes care of the imbalanced distribution of samples in the training set. The proposed model achieves 7.63% improvement in terms of accuracy with respect to the best so far related model for automatic speech recognition. • It provides an online tool for Automatic Speech Recognition for Portuguese language. • An ensemble of expert neural networks is explored, dividing the decision. • Expert only takes care of a well-delimited area of this decision space. • Two models are proposed: (1) grouping similar phonetics; (2) unbalanced dataset. • A dynamic post-processing using an RNN is performed, mitigating of output oscillatory effects. [ABSTRACT FROM AUTHOR]

Abstract: There are several neural network implementations using either software, hardware-based or a hardware/software co-design. This work proposes a hardware architecture to implement an artificial neural network (ANN), whose topology is the multilayer perceptron (MLP). In this paper, we explore the parallelism of neural networks and allow on-the-fly changes of the number of inputs, number of layers and number of neurons per layer of the net. This reconfigurability characteristic permits that any application of ANNs may be implemented using the proposed hardware. In order to reduce the processing time that is spent in arithmetic computation, a real number is represented using a fraction of integers. In this way, the arithmetic is limited to integer operations, performed by fast combinational circuits. A simple state machine is required to control sums and products of fractions. Sigmoid is used as the activation function in the proposed implementation. It is approximated by polynomials, whose underlying computation requires only sums and products. A theorem is introduced and proven so as to cover the arithmetic strategy of the computation of the activation function. Thus, the arithmetic circuitry used to implement the neuron weighted sum is reused for computing the sigmoid. This resource sharing decreased drastically the total area of the system. After modeling and simulation for functionality validation, the proposed architecture synthesized using reconfigurable hardware. The results are promising. [Copyright &y& Elsevier]