毕业生知识库

In recent years, tremendous progress has been made in the field of human-computer interaction. The human hand is one of the human parts interacting most frequently and naturally with the outside world. Therefore, the human-computer interaction research based on human hand motion has received extensive attention, and the hand tracking algorithm is the basis and key to its realization. The motion of the human hand has the characteristics of high speed and variable acceleration. Therefore, as a highly maneuvering target, the human hand poses a challenge to the position tracking algorithm. The nonlinear Kalman filter algorithm is an effective solution; however, it cannot provide real-time estimation accuracy feedback due to the lack of consistency, and the credibility of the algorithm is limited. Therefore, in this paper, the study of hand tracking algorithm based on consistent Kalman filter is carried out.

The main work and innovations of this paper are summarized as follows:

1. To address the problem that the hand tracking algorithm lacks the real-time feed[1]back of estimation accuracy, the hand tracking algorithm based on consistent Kalman filter is proposed. According to the complex and variable acceleration characteristics of the hand motion, the hand motion model considering unknown nonlinearity and uncertain dynamics is established. On the basis of this model, with reasonable assumptions and theoretical proofs, the hand tracking algorithm based on consistent extended Kalman filter is proposed. In the actual experiment, the maximum values of the root mean square error of the position estimation component of the proposed algorithm, the classic extended Kalman filter algorithm and the classic unscented Kalman filter algorithm are 2.06mm, 53.59mm and 24.88mm, respectively; the minimum values of the mean of the bounded percentage are 97.44%, 26.09% and 6.74%, respectively. The experimental results show that the proposed algorithm has good tracking performance and can provide real-time feedback of the estimation accuracy.

2. To address the parameter optimization problem of the proposed hand tracking algorithm, the parameter optimization methods to “minimize the sum of the root mean square errors of the estimated components” are proposed. In order to quantify the estimation performance of the algorithms, the sum of the root mean square errors of the estimated components is chosen as the overall performance index of the algorithms. Aiming at this index, three parameter optimization methods of the hand tracking algorithm based on genetic algorithm, particle swarm optimization algorithm and their hybrid algorithm are proposed. In the simulation and the actual experiment, the minimum performance improvements of the optimized algorithms are 2.84% and 13.62%, respectively, which verifies the effectiveness of the parameter optimization methods.

3. Using the proposed hand tracking algorithm, a preliminary human-computer interaction study for motion assessment is carried out. A human-computer interaction platform for motion assessment is built by Unity 3D software, and the tracking of hand motion is realized by the Leap Motion controller and the algorithm proposed in this paper. The preliminary experiments for motion assessment are carried out through this human-computer interaction platform, which provides a preliminary reference for the subsequent research on motion assessment based on hand tracking algorithm.