This article presents the software and hardware implementation of a low cost and high performance image edge detection algorithm. This algorithm will be used as part of a complete vision based driver assistance system. The main challenge consists in realizing a real-time implementation of edge detection algorithm that contributes in increasing the performance of the whole system. The software implementation of the developed algorithm using MATLAB tool is discussed in this paper, as well as the hardware architecture developed using VHDL language. Test results for both implementations were presented and compared to other edge detection operators. Computational time and other features comparison have shown the effectiveness of the proposed approach. General TermsComputer science, Image processing, VHDL

This paper discusses the evaluation of two supervised learning based image classification algorithms. The classification subject of this work is part of a complete vision based road sign recognition system to be implanted using the VHDL language on an FPGA card for driver assistance applications. The classification is used in order to classify road scene images into different day times according to scene illumination and weather conditions. Due to the sensitivity of colors to illumination variation, the classification task is developed to improve the red color segmentation task which presents an important level in the road sign recognition system. In order to achieve real-time processing tasks and to reduce computing time and hardware resources occupation, the performance of the two predictive modeling techniques which are Neural Networks and Decision Trees is evaluated in this work. The VHDL circuit of the Decision Tree classifier is presented as well.

Abstract-This Paper presents a new hybrid technique for digit recognition applied to the speed limit sign recognition task. The complete recognition system consists in the detection and recognition of the speed signs in RGB images. A pretreatment is applied to extract the pictogram from a detected circular road sign, and then the task discussed in this work is employed to recognize digit candidates. To realize a compromise between performances, reduced execution time and optimized memory resources, the developed method is based on a conjoint use of a Neural Network and a Decision Tree. A simple Network is employed firstly to classify the extracted candidates into three classes and secondly a small Decision Tree is charged to determine the exact information. This combination is used to reduce the size of the Network as well as the memory resources utilization. The evaluation of the technique and the comparison with existent methods show the effectiveness.

No abstract

Purpose The purpose of this paper is to design and validate an electronic nose (E-nose) prototype using commercially available metal oxide gas sensors (MOX). This prototype has a sensor array board that integrates eight different MOX gas sensors to handle multi-purpose applications. The number of sensors can be adapted to match different requirements and classification cases. The paper presents the validation of this E-nose prototype when used to identify three gas samples, namely, alcohol, butane and cigarette smoke. At the same time, it discusses the discriminative abilities of the prototype for the identification of alcohol, acetone and a mixture of them. In this respect, the selection of the appropriate type and number of gas sensors, as well as obtaining excellent discriminative abilities with a miniaturized design and minimal computation time, are all drivers for such implementation. Design/methodology/approach The suggested prototype contains two main parts: hardware (low-cost components) and software (Machine Learning). An interconnection printed circuit board, a Raspberry Pi and a sensor chamber with the sensor array board make up the first part. Eight sensors were put to the test to see how effective and feasible they were for the classification task at hand, and then the bare minimum of sensors was chosen. The second part consists of machine learning algorithms designed to ensure data acquisition and processing. These algorithms include feature extraction, dimensionality reduction and classification. To perform the classification task, two features taken from the sensors’ transient response were used. Findings Results reveal that the system presents high discriminative ability. The K-nearest neighbor (KNN) and support vector machine radial basis function based (SVM-RBF) classifiers both achieved 97.81% and 98.44% mean accuracy, respectively. These results were obtained after data dimensionality reduction using linear discriminant analysis, which is more effective in terms of discrimination power than principal component analysis. A repeated stratified K-cross validation was used to train and test five different machine learning classifiers. The classifiers were each tested on sets of data to determine their accuracy. The SVM-RBF model had high, stable and consistent accuracy over many repeats and different data splits. The total execution time for detection and identification is about 10 s. Originality/value Using information extracted from transient response of the sensors, the system proved to be able to accurately classify the gas types only in three out of the eight MQ-X gas sensors. The training and validation results of the SVM-RBF classifier show a good bias-variance trade-off. This proves that the two transient features are sufficiently efficient for this classification purpose. Moreover, all data processing tasks are performed by the Raspberry Pi, which shows real-time data processing with miniaturized architecture and low prices.