Sergio Ledesma-Orozco

Eigen-Gradients for Traffic Sign Recognition

Traffic sign detection and recognition systems include a variety of applications like autonomous driving, road sign inventory, and driver support systems. Machine learning algorithms provide useful tools for traffic sign identification tasks. However, classification algorithms depend on the preprocessing stage to obtain high accuracy rates. This paper proposes a road sign characterization method based on oriented gradient maps and the Karhunen-Loeve transform in order to improve classification performance. Dimensionality reduction may be important for portable applications on resource constrained devices like FPGAs; therefore, our approach focuses on achieving a good classification accuracy by using a reduced amount of attributes compared to some state-of-the-art methods. The proposed method was tested using German Traffic Sign Recognition Benchmark, reaching a dimensionality reduction of 99.3% and a classification accuracy of 95.9% with a Multi-Layer Perceptron.

Boosting for Image Interpretation by Using Natural Features

In this paper a research in classification of natural images by using Adaboost (adapting boosting) method is presented. This technique is used to identify the nature of the main regions in the image, that is, to identify if they are roads, trees, shades, sky, bushes or others interesting regions; image is previously segmented and each of its regions are represented by a R12 data vector (including features as color, texture and context), in at least 5 classes. The proposed methodology is presented for a multi-class classification problem and for validating our results, performances ratios between Adaboost and the support vector machines are discussed. This methodology is intent to be applied in medical imagery and in visual based navigation on natural environments; in robot navigation, very good results are obtained even in poorly color saturated images. Finally, the results are described and presented showing that Adaboost is a reliable classification technique giving slightly better performances than SVM for regions in natural images.

New prioritized value iteration for Markov decision processes

The problem of solving large Markov decision processes accurately and quickly is challenging. Since the computational effort incurred is considerable, current research focuses on finding superior acceleration techniques. For instance, the convergence properties of current solution methods depend, to a great extent, on the order of backup operations. On one hand, algorithms such as topological sorting are able to find good orderings but their overhead is usually high. On the other hand, shortest path methods, such as Dijkstra’s algorithm which is based on priority queues, have been applied successfully to the solution of deterministic shortest-path Markov decision processes. Here, we propose an improved value iteration algorithm based on Dijkstra’s algorithm for solving shortest path Markov decision processes. The experimental results on a stochastic shortest-path problem show the feasibility of our approach.

Diagnosis test of power cables using a resonant test system and analysis of partial discharge on site

To after-laying of new-installed high voltage (HV) power cables the use of on-site non-destructive on-site testing and diagnosis of new installations of the power cables is becoming an important issue to determine the actual condition of the cable systems and to determine the future performances. An overview is presented on on-site testing using a resonant test system and partial discharge diagnosis of HV power cables with regard to on-site testing methods: AC voltage test of the insulation, diagnostic aspects using partial discharge measurement, patterns recognition using self organizing map (SOM).

Tank Model Coupled with an Artificial Neural Network

Tank models have been used in many Asian countries for flood forecasting, reservoir operation, river basin modeling, etc. In this work a tank model is coupled with an ANN (Artificial Neural Network) for modeling a rainfall-runoff process. The ANN controls six of the tank model parameters to adjust them along time in order to improve efficiency. The data used in the simulations were collected from the Brue catchment in the South West of England. It should be pointed out that the raingauge network in this study is extremely dense (for research purposes) and does not represent the usual raingauge density in operational flood forecasting systems.

Applying balancing techniques in traffic sign recognition

Traffic Sign Recognition systems aim to determine the meaning of traffic signs in highways for real-world applications such astraffic sign inventory or driver assistance systems. Traffic sign datasets are inherently imbalanced, i.e. some traffic signs appearmore frequently than others. One serious consequence of this imbalance is the low recognition rates of minority classes (classeswith fewer training cases). In this paper, we propose a new method for improving traffic sign recognition of minority classes, byapplying balancing algorithms. As a result, our proposed method improves minority class recognition rates up to 28% comparedto traditional methods.

Traffic Sign Recognition Based on Linear Discriminant Analysis

Traffic Signs provide visual information to drivers, in order to warn them from possible danger on the road, set rules for pedestrian protection and inform people about their environment, to name a few. Therefore, Traffic Sign Detection and Recognition Systems have increased their interest in the scientific community. Applications include autonomous driving systems, road sign inventory and driver support assistance systems. This paper presents a traffic sign recognition algorithm for velocity signs, based on Linear Discriminant Analysis that performs dimensionality reduction and it improves class separability. The tests were performed on the German Traffic Sign Recognition Benchmark, using a Multi-Layer Perceptron as a classification tool. LDA classification and k-Nearest Neighbors were also used for comparison. Experimental results demonstrate the validity of the proposed approach, having a 99.1% of attributes reduction and a 96.5% of classification accuracy.

Reduction of temporal complexity in Markov decision processes

In this paper we propose a new algorithm in order to reduce temporal complexity in Markov decision processes. Value iteration is a classical algorithm for solving them, but this algorithm and its variants are quite slow for discount factors close to one and their convergence properties depend to a great extent on a good state update order. It has been shown that improved topological value iteration presents a good convergence speed thanks to the use of an improved topological ordering. Nevertheless, its drawback is due to high memory requirements. So, our algorithm obtains the optimal state backup order with less memory requirements. Experimental results on stochastic shortest-path problems (highly cyclic) are presented. Our approach obtained a considerable reduction in temporal complexity with respect to other variants of value iteration. For instance, the experiments showed in one test a reduction of six times with respect to asynchronous value iteration.

Combination of acceleration procedures for solving stochastic shortest-path Markov decision processes

In this paper we propose the combination of accelerated variants of value iteration with improved prioritized sweeping for the solution of stochastic shortest path Markov decision processes. For the fastest solution, asynchronous updates, prioritization and prioritized sweeping have been tested. A topological reordering algorithm was also compared with a static reordering algorithm. Experimental results obtained on afinite state and action-space stochastic shortest path problem are presented.