An Interval Type-2 Fuzzy Logic System for Human Silhouette Extraction in Dynamic Environments

In this paper, we present a type-2 fuzzy logic based system for robustly extracting the human silhouette which is a fundamental and important procedure for advanced video processing applications, such as pedestrian tracking, human activity analysis and event detection. The presented interval type-2 fuzzy logic system is able to detach moving objects from extracted human silhouette in dynamic environments. Our real-world experimental results demonstrate that the proposed interval type-2 fuzzy logic system works effectively and efficiently for moving objects detachment where the type-2 approach outperforms the type-1 fuzzy system while significantly reducing the misclassification when compared to the type-1 fuzzy system. K eywords : Interval type 2 fuzzy logic, Silhouette extraction, Human tracking. 1 IN T R O DU C T I O N Accurate human silhouette (or outline) segmentation from a video sequence is important and fundamental for advanced video applications such as pedestrian tracking and recognition, human activity analysis and event detection. Advanced human detection and identification approaches like [1], [2] can be utilized for silhouette extraction. However, such methods are commonly of high computational complexity and hence not suitable for dynamic and complex environments. Hence, there is a need for silhouette extraction methods which are computationally efficient and that are able to operate in dynamic and complex environments. The first step for silhouette extraction is background modeling and subtraction to detect moving targets as foreground objects. In [3], an approach based on a single Gaussian modal was developed which employed a simple robust method to handle moving objects and slow illumination changes. However, there are several limitations in this method such as learning stage necessity for background distribution, and robustness deficiency for situations like sudden illumination changes, slow moving objects, etc. To address these problems, the Gaussian Mixture Model (GMM) was proposed [4]. In this model, each pixel is modeled using n Gaussian distributions. GMM is effective to overcome the shortcomings of single Gaussian model and hence GMM is extensively recognized as a robust approach for background modeling and subtraction. Therefore, in this paper, GMM is utilized for foreground detection. However, it is unreasonable to simply consider GMM foreground as human silhouette in real-life environments because there are numerous noise factors and uncertainties to handle which include: Varying light condition Reflections and shadows Moving objects attached to human silhouette (a book, a chair, etc.). To handle these problems and detach the moving objects from the human silhouette, a type-1 Fuzzy Logic System (T1FLS) was proposed [5]. This T1FLS is capable of handling to an extent the uncertainties mentioned above, however, the extracted silhouette will be degraded due to misclassification of the proposed T1FLS. Hence, in this paper, we will present an Interval Type-2 Fuzzy Logic System (IT2FLS) which will be able to handle the high uncertainty levels present in real-world dynamic environments while also reducing the misclassification of extracted silhouette. The IT2FLS used similar type-1 membership function as the ones presented in [5] as principal membership functions which are then blurred to produce the type-2 fuzzy sets used in this paper. We have also used the same rule base as [5] to allow for a fair comparison with the results reported in [5]. In this proposed system, GMM is adopted for original foreground detection, then a IT2FLS is performed to detach the moving objects from the human silhouette. We have performed several real-world experiments where it was shown that the proposed IT2FLS is effective to reduce the misclassification and the quality of the extracted human silhouette is much improved when compared to the T1FLS. The rest of this paper is organized as follows. In section 2, we provide a brief overview of type-2 FLSs. Section 3 presents the proposed IT2FLS. Section 4 presents the experiments and results and finally the conclusions and future work are presented in section 5. F ig.1.(a) Structure of the type-2 FLS. (b) An interval type-2 fuzzy set 2 A BRI E F O V E R V I E W O F T H E I T2F LS The IT2FLS depicted in Fig. 1a) [6] uses interval type-2 fuzzy sets (such as the type-2 fuzzy set shown in Fig. 1b) [6] to represent the inputs and/or outputs of the FLS. In the interval type-2 fuzzy sets all the third dimension values are equal to one [6], [7]. The use of interval type-2 FLS helps to simplify the computation (as opposed to the general type-2 FLS) [8]. The interval type-2 FLS works as follows [6], [7], [8]: the crisp inputs from the input sensors are first fuzzified into input type-2 fuzzy sets; singleton fuzzification is usually used in interval type-2 FLS applications due to its simplicity and suitability for embedded processors and real time applications. The input type-2 fuzzy sets then activate the inference engine and the rule base to produce output type-2 fuzzy sets. The type-2 FLS rule base remains the same as for the type-1 FLS but its Membership Functions (MFs) are represented by interval type-2 fuzzy sets instead of type-1 fuzzy sets. The inference engine combines the fired rules and gives a mapping from input type-2 fuzzy sets to output type-2 fuzzy sets. The type-2 fuzzy output sets of the inference engine are then processed by the type-reducer which combines the output sets and performs a centroid calculation which leads to type-1 fuzzy sets called the typereduced sets. There are different types of type-reduction methods. In this paper we will be using the Centre of Sets type-reduction as it has reasonable computational complexity that lies between the computationally expensive centroid type-reduction and the simple height and modified height type-reductions which have problems when only one rule fires [6], [7]. After the type-reduction process, the type-reduced sets are defuzzified (by taking the average of the type-reduced set) to obtain crisp outputs that are sent to the actuators. More information about the interval type-2 FLS and its benefits can be found in [6], [7], [8]. 3 T H E PR OPOSE D I T2F LS F O R H U M A N SI L H O U E T T E