Moving object tracking in the low-rank representation

Abstract

An artificial olfactory system, referred to an electronic nose, is a multi-sensor platform used for gas classification. Lack of selectivity and low repeatability of the gas sensors are the major challenges in all gas identification problems. Pattern recognition algorithms are combined with a sensor array to address these challenges. The implementation of these algorithms is another challenge for the hardware friendly system. In this paper, we introduce a hardware friendly algorithm for gas identification. In this algorithm, we use sensitivity difference of any two sensors in the array as an input feature and a subset of the features is extracted by evaluating the capability of each pair of sensor to split the gases into two branches. The learning process of the pairs of sensors continues at every split point on the way until all individual gases are identified. The learned pairs of sensors at each split point are used for the identification of a new test response pattern and plurality voting is used for the distribution of the gases in cases of contention among the pairs. In order to assess the performance of our approach, a 4x4 tin-oxide gas sensor array is used to acquire the data of three gases in a laboratory. Accuracy rate of 100% is achieved with our algorithm on this experimental data set.