Miguel A. Molina-Cabello

Smart motion detection sensor based on video processing using self-organizing maps

A low cost smart motion detector is presented.It is based on the Arduino DUE microcontroller.The software architecture employs a fixed point arithmetic paradigm.The self-organizing map neural network is implemented on chip.The performance is substantially higher than that of the traditional detector. Most current approaches to computer vision are based on expensive, high performance hardware to meet the heavy computational requirements of the employed algorithms. These system architectures are severely limited in their practical application due to financial and technical limitations. In this work a different strategy is used, namely the development of an inexpensive and easy to deploy computer vision system for motion detection. This is achieved by three means. First of all, an affordable and flexible hardware platform is employed. Secondly, the motion detection algorithm is specifically tailored to involve a very small computational load. Thirdly, a fixed point programming paradigm is followed in implementing the system so as to further reduce the computational requirements. The proposed system is experimentally compared to the standard motion detector for a wide range of benchmark videos. The reported results indicate that our proposal attains substantially better performance, while it remains affordable and easy to install in practice.

Vehicle Type Detection by Convolutional Neural Networks

In this work a new vehicle type detection procedure for traffic surveillance videos is proposed. A Convolutional Neural Network is integrated into a vehicle tracking system in order to accomplish this task. Solutions for vehicle overlapping, differing vehicle sizes and poor spatial resolution are presented. The system is tested on well known benchmarks, and multiclass recognition performance results are reported. Our proposal is shown to attain good results over a wide range of difficult situations.

The effect of noise on foreground detection algorithms

Background segmentation methods are exposed to the effects of different kinds of noise due to the limitations of image acquisition devices. This type of distortion can worsen the performance of segmentation methods because the input pixel values are altered. In this paper we study how several well-known background segmentation methods perform when the input is corrupted with several levels of uniform and Gaussian noise. Furthermore, few situations are reported where instead of an inconvenience, adding noise to the input may be desirable to attenuate some limitations of a method. In this work, the performance of nine well known methods is studied under both kinds of noise.

Panoramic background modeling for PTZ cameras with competitive learning neural networks

The construction of a model of the background of a scene still remains as a challenging task in video surveillance systems, in particular for moving cameras. This work presents a novel approach for constructing a panoramic background model based on competitive learning neural networks and a subsequent piecewise linear interpolation by Delaunay triangulation. The approach can handle arbitrary camera directions and zooms for a Pan-Tilt-Zoom (PTZ) camera-based surveillance system. After testing the proposed approach on several indoor sequences, the results demonstrate that the proposed method is effective and suitable to use for real-time video surveillance applications.

Neural controller for PTZ cameras based on nonpanoramic foreground detection

In this paper a controller for PTZ cameras based on an unsupervised neural network model is presented. It takes advantage of the foreground mask generated by a non-parametric foreground detection subsystem. Thus, our aim is to optimize the movements of the PTZ camera to attain the maximum coverage of the observed scene in presence of moving objects. A growing neural gas (GNG) is applied to enhance the representation of the foreground objects. Both qualitative and quantitative results are reported using several widely used datasets, which demonstrate the suitability of our approach.

Foreground Detection by Competitive Learning for Varying Input Distributions

One of the most important challenges in computer vision applications is the background modeling, especially when the background is dynamic and the input distribution might not be stationary, i.e. the distribution of the input data could change with time (e.g. changing illuminations, waving trees, water, etc.). In this work, an unsupervised learning neural network is proposed which is able to cope with progressive changes in the input distribution. It is based on a dual learning mechanism which manages the changes of the input distribution separately from the cluster detection. The proposal is adequate for scenes where the background varies slowly. The performance of the method is tested against several state-of-the-art foreground detectors both quantitatively and qualitatively, with favorable results.

Pixel Features for Self-organizing Map Based Detection of Foreground Objects in Dynamic Environments

Among current foreground detection algorithms for video sequences, methods based on self-organizing maps are obtaining a greater relevance. In this work we propose a probabilistic self-organising map based model, which uses a uniform distribution to represent the foreground. A suitable set of characteristic pixel features is chosen to train the probabilistic model. Our approach has been compared to some competing methods on a test set of benchmark videos, with favorable results.

Blood Cell Classification Using the Hough Transform and Convolutional Neural Networks

The detection of red blood cells in blood samples can be crucial for the disease detection in its early stages. The use of image processing techniques can accelerate and improve the effectiveness and efficiency of this detection. In this work, the use of the Circle Hough transform for cell detection and artificial neural networks for their identification as a red blood cell is proposed. Specifically, the application of neural networks (MLP) as a standard classification technique with (MLP) is compared with new proposals related to deep learning such as convolutional neural networks (CNNs). The different experiments carried out reveal the high classification ratio and show promising results after the application of the CNNs.

Foreground Detection Enhancement Using Pearson Correlation Filtering

Foreground detection algorithms are commonly employed as an initial module in video processing pipelines for automated surveillance. The resulting masks produced by these algorithms are usually postprocessed in order to improve their quality. In this work, a postprocessing filter based on the Pearson correlation among the pixels in a neighborhood of the pixel at hand is proposed. The flow of information among pixels is controlled by the correlation that exists among them. This way, the filtering performance is enhanced with respect to some state of the art proposals, as demonstrated with a selection of benchmark videos.

Background Modeling for Video Sequences by Stacked Denoising Autoencoders

Nowadays, the analysis and extraction of relevant information in visual data flows is of paramount importance. These images sequences can last for hours, which implies that the model must adapt to all kinds of circumstances so that the performance of the system does not decay over time. In this paper we propose a methodology for background modeling and foreground detection, whose main characteristic is its robustness against stationary noise. Thus, stacked denoising autoencoders are applied to generate a set of robust characteristics for each region or patch of the image, which will be the input of a probabilistic model to determine if that region is background or foreground. The evaluation of a set of heterogeneous sequences results in that, although our proposal is similar to the classical methods existing in the literature, the inclusion of noise in these sequences causes drastic performance drops in the competing methods, while in our case the performance stays or falls slightly.