Computer Visual Based Object Recognition And Tracking Algorithm For HCI

With the development of human-computer interaction (HCI) technology, more intelligent, more natural human-computer interaction will become the trend of development in the future. As a natural and intuitive interactive way, gestures play an increasingly important role in HCI. Computer vision-based gesture recognition is the next generation of HCI, has important theoretical research value and application prospects.Normally, a computer vision based gesture recognition system has three parts, hand the detecting and tracking, static gesture recognition, and dynamic gesture recognition. This article has studied several key technologies based on monocular vision hand gesture recognition technology, including image preprocessing, skin color segmentation, tracking and recognition algorithm.This article focuses on computer vision-based target tracking algorithm. In the research of the Mean Shift and CAMSHIFT tracking algorithm, issued an improved CAMSHIFT algorithm. This algorithm may resume the track through the automatic expanded search window's method after the tracking object loss. The algorithm operation is simple, timeliness is high, suits the real-time track application. Moreover, started by the Baye filter principle, research the Kalman filter, the granule filter and the particle filtering theory in the visual track application.The article has designed a model based and apparent based hand gesture recognition system. The proposed approach uses a set of 2D hand models in place of high degree-of-freedom and high-dimensional 3D model, the model suit the distortion hand gesture recognition. Based on this model, the article proposed a new method of curvature based fingertip detection. In the system, gesture tracking uses the Kalman filtering and an improved CAMSHIFT algorithm combining locate gesture region, this algorithm can search through the Kalman filter to estimate the window with the gestures and the color correction to resolve the impact of regional overlap, and also can solve the gesture movement to accelerate or move out of camera view to track the gestures resulting from the failure of the problem with improved CAMSHIFT algorithm. After using the YCbCr color space skin detection technology for gesture segmentation, and finger gesture contour detection, using the key contour to track fingers with particle filter, reducing the computational algorithm. This method is applied to a stable, real-time gesture tracking applications that can be deformed to meet the real-time vision-based human-computer interaction requirements.