Vasile Gui

Bio: Vasile Gui is an academic researcher from Politehnica University of Timișoara. The author has contributed to research in topics: Mean-shift & Image segmentation. The author has an hindex of 11, co-authored 61 publications receiving 367 citations.

A Sensor Network Approach for Violence Detection in Smart Cities Using Deep Learning.

Abstract: Citizen safety in modern urban environments is an important aspect of life quality. Implementation of a smart city approach to video surveillance depends heavily on the capability of gathering and processing huge amounts of live urban data. Analyzing data from high bandwidth surveillance video streams provided by large size distributed sensor networks is particularly challenging. We propose here an efficient method for automatic violent behavior detection designed for video sensor networks. Known solutions to real-time violence detection are not suitable for implementation in a resource-constrained environment due to the high processing power requirements. Our algorithm achieves real-time processing on a Raspberry PI-embedded architecture. To ensure separation of temporal and spatial information processing we employ a computationally effective cascaded approach. It consists of a deep neural network followed by a time domain classifier. In contrast with current approaches, the deep neural network input is fed exclusively with motion vector features extracted directly from the MPEG encoded video stream. As proven by results, we achieve state-of-the-art performance, while running on a low computational resources embedded architecture.

A fast algorithm for background tracking in video surveillance, using nonparametric kernel density estimation

Abstract: Moving object detection and tracking in video surveillance systems is commonly based on background estimation and subtraction. For satisfactory performance in real world applications, robust estimators, tolerating the presence of outliers in the data, are needed. Nonparametric kernel density estimation has been successfully used in modeling the background statistics due to its capability to perform well without making any assumption about the form of the underlying distributions. However, in real-time applications, the O(N2) complexity of the method can be a bottleneck preventing the object tracking and event analysis modules from having the computing time needed. In this paper, we propose a new background subtraction technique, using multiresolution and recursive density estimation with mean shift based mode tracking. An algorithm with complexity independent on N is developed for fast, real-time implementation. Comparative results with known methods are included, in order to attest the effectiveness and quality of the proposed approach.

Computer aided diagnosis method for steatosis rating in ultrasound images using random forests

Abstract: In this paper we discuss the problem of computer aided evaluation of the severity of steatosis disease using ultrasound images, the aim of the study being to compare the automatic evaluation of liver steatosis using random forests (RF) and support vector machine (SVM) classifiers. Material and method: One hundred and twenty consecutive patients with steatosis or normal liver, assessed by ultrasound by the same expert, were enrolled. We graded steatosis in four stages and trained two classifiers to rate the severity of disease, based on a large set of labeled images and a large set of features, including several features obtained by robust estimation techniques. We compared RF and SVM classifiers. The classifiers were trained using cross-validation. There was 80% of data randomly selected for training and 20% for testing the classifier. This procedure was performed 20 times. The main measure of performance was the accuracy. Results: From all cases, 10 were rated as normal liver, 70 as hav ing mild, 33 moderate, and 7 severe steatosis. Our best experts’ ratings were used as ground truth data. RF outperformed the SVM classifier and confirmed the ability of this classifier to perform well without feature selection. In contrast, the performance of the SVM classifier was poor without feature selection and improved significantly after feature selection. Conclusion: The ability and accuracy of RF to classify well the steatosis severity, without feature selection, were superior as compared to SVM.

Multivariate Density Estimation

Abstract: Exploring and identifying structure is even more important for multivariate data than univariate data, given the difficulties in graphically presenting multivariate data and the comparative lack of parametric models to represent it. Unfortunately, such exploration is also inherently more difficult.

A Local Contrast Method for Small Infrared Target Detection

Abstract: Robust small target detection of low signal-to-noise ratio (SNR) is very important in infrared search and track applications for self-defense or attacks. Consequently, an effective small target detection algorithm inspired by the contrast mechanism of human vision system and derived kernel model is presented in this paper. At the first stage, the local contrast map of the input image is obtained using the proposed local contrast measure which measures the dissimilarity between the current location and its neighborhoods. In this way, target signal enhancement and background clutter suppression are achieved simultaneously. At the second stage, an adaptive threshold is adopted to segment the target. The experiments on two sequences have validated the detection capability of the proposed target detection method. Experimental evaluation results show that our method is simple and effective with respect to detection accuracy. In particular, the proposed method can improve the SNR of the image significantly.