Ming Zeng

Mobile robot gas source localization via top-down visual attention mechanism and shape analysis

A novel mobile robot based gas source localization method in which the top-down visual attention mechanism (TDVAM) is combined with shape analysis is proposed. At each location, three different images which cover the scene in front of the robot are captured via changing the horizontal angle of an onboard pan/tilt camera. In each image, three salient regions are computed using TDVAM model, and maximal one plausible gas source from three salient regions is identified by using the shape analysis. The positions of recognized plausible gas sources are determined with a laser range scanner. The robot is navigated to the plausible gas sources one by one and gas concentration information is used to judge if the plausible sources are real ones. Experimental results in a complex indoor airflow environment in which valves are used to simulate gas sources demonstrate the feasibility and high efficiency of the proposed gas source localization approach.

A New Method Combining KECA-LDA With ELM for Classification of Chinese Liquors Using Electronic Nose

We proposed a hybrid algorithm by combining kernel entropy component analysis (KECA) with linear discriminant analysis (LDA), namely, KECA-LDA for feature reduction in electronic-nose systems. It combined the advantages of KECA and LDA. Then, the data extracted by KECA-LDA were inputted to extreme learning machine (ELM) for classification. In order to examine the performance of the proposed method, eight types of strong-flavor Chinese liquors were tested using an electronic nose (e-nose) system designed by ourselves, and the results after cross validation showed that features extracted by KECA-LDA were more beneficial to classification than KECA, and the performance of ELM was better than that of backpropagation neural network. The highest classification rate by KECA-LDA-ELM was 100%. In conclusion, an e-nose combined with KECA-LDA and ELM is a feasible method to classify Chinese liquors.

Mobile robot based odor source localization via particle filter

We consider odor-source localization using a mobile robot in a time-variant airflow-field environment. Novel plume tracing and odor-source declaration methods are presented. When odor plume clue is found, an odor-patch path is estimated by a dynamic-window approach, and the robot traces the plume along a route planned from the odor-patch path. In parallel, a particle filter is used to localize an odor source. The source is declared if the estimated locations converge in a relatively small area for a given period. In view of the common foundational odor concentration that already exists in the local or even whole searched area before the robot searches, differential concentration based on moving-average value is used to obtain an adaptively variable concentration threshold. Experiment results in an indoor time-variant airflow experiment show that the robot can effectively approach and declare the odor source.

Multi-robot odor-plume tracing in indoor natural airflow environments using an improved ACO algorithm

We consider odor-plume tracing using a multi-robot system in indoor natural airflow environments. The purpose of odor-plume tracing is to approach the odor source via following the found plume with mobile robot(s). Owing to the chaotic nature of the odor transport in the atmosphere, tracing the resultant patchy meandering plume down to its source is thus not a trivial task. A novel multi-robot based plume tracing method is proposed. When the plume is found, multiple robots are coordinated to trace the time-variant plume by an improved ant colony optimization (ACO) algorithm combined with upwind search. The results of experiments compared with the spiral surge approach using real robots in indoor natural airflow environments validate the feasibility and robustness of the proposed plume tracing method.

Mobile Robots Odor Localization with an Improved Ant Colony Algorithm

An improved ant colony algorithm (ACA) is put forward to solve the mobile robot odor localization (MROL) problem. The so-called MROL means localizing an odor source with mobile robots. The improved algorithm is realized through three phases, which are genetic algorithm (GA) based local search, global search and pheromone update. The GA ensures that the optimal or sub-optimal points can be found within local areas. The global search phase consists of random and probability based searches. The random search can prevent the ACA from getting into local optimum. Detailed implementation procedure of the improved ACA for the MROL is presented. Two Gaussian concentration models are used to describe the odor distribution. Simulation results show that the robots can asymptotically approach and finally determine the odor source.

Nonlinear analysis of the near-surface wind speed time series

Research on characteristics of the near-surface wind speed time series can be of great help to understand the mechanisms of odor/gas dispersal and moreover provides useful clues for the optimization of odor/gas source localization algorithms. In this paper, an integrated technique which combines a direct identification method of chaos, i.e., the saturated correlation dimension algorithm with the surrogate data method (an indirect identification method) is proposed to analyze the chaotic characteristics of the near-surface wind speed time series. The analysis procedure includes two stages. Firstly, the GP algorithm is applied to calculate the saturated correlation dimension of the wind speed time series. The value of correlation dimension is 4.1628 ± 0.0022, which indicates that the wind speed time series probably has chaotic characteristics. Then 30 surrogate data sets of original wind speed series are generated by amplitude adjusted Fourier transform (AAFT), and saturated correlation dimension is employed as the statistic of Sigma test. Simulation experiments and numerical analysis show that some deterministic nonlinear components exist in the original wind speed time series. This conclusion provides further confirmation to our pre-assumption, i.e. the near-surface wind speed signal may have chaotic characteristics.

Motion capture and reconstruction based on depth information using Kinect

In this paper, a novel technique which provides visual feedback to the trainee in a 3D virtual environment is proposed. This method contains two steps. Firstly, the real-time depth data for the 3D human motions are captured using Kinect (a latest depth sensor launched by Microsoft) and then those depth dada are converted into key-node data of human skeletons. Next, 3D human body movements are reconstructed by combining these key-node data and personalized virtual models of human body created by the open source software of “MakeHuman”. The experimental results show that the proposed algorithm can obtain a fairly accurate estimation of the real-time 3D human body movements.

Single odor source declaration in outdoor time-variant airflow environments

This paper addresses the problem of single odor source declaration in outdoor time-variant airflow environments using a mobile robot. The airflow velocity (including the magnitude and direction) changes fast in outdoor environments, so it is not easy to determine whether a possible source is the true source or not. To identify the possible source under the uncertainty caused by the chaotic dispersion of the odor, a logic rule based on statistics is proposed, and a circular route surrounding the possible source is planned for the robot to collect the airflow velocities and odor concentrations. The possible source is declared when the circular motion is complete. Experiment results show that the possible source can be identified with a high success rate by using the proposed method.

Tracing Odor Plume by Robot in Time-variant Flow-field Environments: Tracing Odor Plume by Robot in Time-variant Flow-field Environments

The reliable plume tracing is a critical issue for the implementation of odor source localization. This paper addresses the problem of tracing plume by a mobile robot in real time-variant flow-field environments. An estimatimation of the most likely path taken by the odor patch is to be done whenever the odor patch is detected by the gas sensor mounted on the mobile robot. Subsequently, a searching path is calculated and followed by the mobile robot to trace the odor plume with consideration of the likely odor path estimated and the flow direction at current time. The relative change of odor concentration is adopted as the concentration field is time-variant and might have a common foundational concentration. The experiment result in the indoor time-variant airflow-field environment has show that the mobile robot can be effectively guided in real time with the proposed method in tracing plume and reaching the odor source.

A Novel Object Recognition Method for Mobile Robot Localizing a Single Odor/Gas Source in Complex Environments

An improved single odor/gas source searching approach using a mobile robot by combining image recognition in complex environments is presented. First, color image segmentation of prospective visual candidates is achieved using support vector machines (SVM). Second, the features of those candidates, such as color, shape and orientation (the posture of the object) are extracted. Third, the robot finds a salient object according to the characteristics of analysis areas. Last, the robot moves towards the object which is the most likely to be an odor/gas source. The robot moves upwind if gas concentration is detected. Otherwise, the robot moves along the new direction obtained from the further analysis. Experimental results show the efficiency and practicality of the approach for localizing a leaking ethanol bottle in complex indoor environments.