Wireless image fuzzy recognition system for human activity

Abstract

This study developed a fuzzy image model system for transmitting data over a wireless network channel to efficiently realize human activity in virtual images presentation. Because of the excellent mobile characteristics of wireless sensing networks, small devices are very desirable for local-area deployment. Complex model identification problems, such as acquiring and handling wireless image patterns, require analyzing a large amount of data, which occupies a long time at an acceptable transmission quality. In the proposed system, a cross-layer access method is employed to improve the visual clarity. Image packets are assigned to tune the category queue priority, with the probability allocated through a Markov chain model. This is a favorable approach to balancing the wireless image transmission traffic load. The similarity mixing algorithm, which is based on the maximal similarity and minimal disparity concepts, is used to aggregate the primary image features. The collected image patterns with converted coding vectors are efficiently trained through a human feature recognition procedure to generate a human model. A human action is received in real time from wireless sensing networks, and the image feature is retrieved by approximating a higher compatibility in practice simulations. The fuzzy image model uses the simple region-based evaluation and flexible extraction concepts to describe appropriate image partitions. This technology provides the highest possibility of human feature maps to identify the current action and offers a simple method for detecting human activity in indoor environments. Several human sensing and feature mapping experiments were conducted to verify the feasibility of applying the image recognition technology in nonlinear, time variant, and uncertain human activity problems. This study integrates numerous advantages from the mobility of wireless sensing; the proposed system efficiently controls congested image packages and easily confirms their related human activity. Experimental results verify that 60 testing frames approach about 96.6% accuracy within 3 s. These evaluations illustrate that it is applicable usage in some indoor environments.