The most prevailing approach now for parking lot vehicle detection system is to use sensor-based techniques such as ultrasound and infrared-light sensors, though surveillance cameras have been installed in many parking lots. Camera-based systems can be used both for security/surveillance and for the management of vehicles. But in practice, the camera-based system is used only for underground and indoor parking lots due to the poor accuracy of the detection system. This paper proposes to improve the performance of the camera-based vehicle detection system by using the classifier based on fuzzy c-means (FCM) clustering. For reducing computation, the semi-hard clustering approach is applied and the hyper-parameters are tuned by particle swarm optimization (PSO). The new system was introduced to an underground parking lot at Ginza, Tokyo in October 2009. The system was also tested at an outdoor (rooftop) parking lot for a period of two months and achieved the detection rate (sensitivity) of 99.6%. The performance clearly surpassed the initial goal of the project.

View full abstract

This paper reports on the performance of the fuzzy c-means based classifier (FCMC) adopting the length of cluster centers and mixing proportions of clusters as the free parameters. The FCMC consists of two phases. The first phase is an unsupervised clustering. The clustering is done on a per class basis and is implemented by using the data from one class at a time. The second phase of FCMC is a supervised classification where the free parameters of the classifier are chosen by particle swarm optimization (PSO). High performance classifiers usually have parameters to be selected. For example, the support vector machine (SVM) has the regularization and kernel parameters. These hyperparameters are chosen by an optimization procedure to improve the generalization ability of the classifiers in terms of cross validation test. The grid search is the popular approach for SVM. Since the FCM classifier has many hyperparameters and the validation set error rate is not a unimodal function of the parameters, for the parameter search, we apply PSO inspired by social behavior of bird flocking or fish schooling. PSO is based on a simple random search and easy-to-implement. UCI benchmark datasets are used to evaluate the performance. FCM classifier in combination with the standard 10-CV procedure for parameter selection achieves better test set performance compared to k-nearest neighbor classifier. The remarkable finding is that the resubstitution (i.e., 1-CV) procedure for parameter selection also shows good test set performance. Randomized test sets performance of the classifier is comparable to that of the support vector machine (SVM) reported in the literature.

View full abstract

The automatic construction of fuzzy system with large number of input variables involves many difficulties such as time complexity and the problem of getting in a local minimum. In order to avoid them, an SIRMs (Single Input Rule Modules) model has been proposed. However, such a simple model does not always show good perfomance in complex non-linear systems. Therefore, we have proposed a fuzzy reasoning model as a generalized SIRMs model, in which each module has small number of input variables. A reasoning output is determined as the weighted sum of all modules, where each weight is the importance degree of a module. In this paper, in order to construct effective model, we introduce the delete functions based the importance degree and forgetting rules to the proposed system. As aresult, learning algorithm to construct a fuzzy reasoning system with small-number-of-input rule modules (SNIRMs) is proposed. The conducted numerical simulations show that the proposed method is superior in constructing effective model to the fomer model.

View full abstract

This paper proposes an extraction method of the subjective optimal solutions that does not burden the decision maker. The method removes unnecessary parts from the existing method based on the multiattribulte utility theory and reduces the number of questions. This paper researches effects on final result from two points of view, “function formula” and “function parameter” of utility function. In the result, function parameter is extracted with difficulty, and it is less affected by final result. Therefore, this paper proposes a new extraction method of utility function, which reduces the number of questions about function parameter. The decision making model in this method expresses the subjectivity of the decision maker by introducing a fuzzy measure, and it realizes evaluation methods between the minimum method and the maximum method and extends the width of subjectivity expression. Therefore the proposed method realizes a decision aid of high precision without heavy burdens to the decision maker.

View full abstract

Aperture problem is a psychological experiment for analyzing binding mechanism of the spatial recognition in an early stage of visual pathway. In this paper, we measure perceptual rate in the aperture experiments, and discuss the dependency between the perception and various parameters in the experiments. We also record Electroencephalograms (EEG) of subjects who are recognizing the perception. By the electroencephalograms (EEG) analysis, we measure reaction latency of visual evoked potential (VEP) and event related potential (ERP) related to visual pathway, and estimate the localized equivalent current dipole (ECD) in the visual pathway.

View full abstract