Vision based hand tracking is an important area of human machine interaction (HCI) and virtual reality techniques. However, it is still a great challenge. Due to the complicated environment and various hand appearances, it is difficult to realize stable and long-term tracking. In this paper, we proposed an effective approach to detect and track hand with a normal webcam. An integrating multi-cue detector with skin color and hand classifier is used to initialize the tracking region and find appearance information during tracking. Also a median flow tracker is integrated which utilizing the motion information to enhance the accuracy in short-term. We realize an automatic system of stable and long-term hand tracking, which can detect and track pre-trained frontal view hand gesture. Experiments show the good performance of our approach.

No abstract

Most works on Support Vector Regression (SVR) focus on kernel or loss functions, with the corresponding support vectors obtained using a fixed-radius [Formula: see text]-tube, affording good predictive performance on datasets. However, the fixed radius limitation prevents the adaptive selection of support vectors according to the data distribution characteristics, compromising the performance of the SVR-based methods. Therefore, this study proposes an “Alterable [Formula: see text]-Support Vector Regression” ([Formula: see text]-SVR) model by applying a novel [Formula: see text], named “Alterable [Formula: see text],” to the SVR model. Based on the data point sparsity at each location, the model solves the different [Formula: see text] at the corresponding position, and thus zoom-in or zoom-out the [Formula: see text]-tube by changing its radius. Such a variable [Formula: see text]-tube strategy diminishes noise and outliers in the dataset, enhancing the prediction performance of the [Formula: see text]-SVR model. Therefore, we suggest a novel non-deterministic algorithm to iteratively solve the complex problem of optimizing [Formula: see text] associated with every location. Extensive experimental results demonstrate that our approach can improve the accuracy and stability on simulated and real data compared with the baseline methods.