Research On 3D Human Body Tracking Algorithms Based On Quantum-Inspired Immune Cloning

Computer vision-based 3D human motion analysis studies how to extract the 2D information of human motion from image sequences which exist human motion, such as locations, motions and gestures, and how to recover the 3D human body posture, which plays an important roles in the wide application such as pose recognition, scene surveillance, medical motion analysis, human-computer interaction, behavior understanding and semantic analysis etc.In recent years, computer vision-based 3D human motion analysis received increasing attention from many scholars domestic and international. Overall computer vision-based 3D human motion analysis is divided into three parts: the human body detection, motion tracking and motion understanding, and the motion tracking is the most important part. The three-dimensional human motion estimation based on video image sequences is a complex high-dimensional nonlinear problem, and also for real-time demanding. For the estimated state vector to be very high dimensional problems in high-dimensional space to conduct a global search or optimization is very time-consuming, but using the deterministic optimization method is very easy to fall into the local optimal solution, and the use of particle filters and other random searching methods are alse not given satisfactory results. In order to improve the complex high-dimensional state space search efficiency, and to obtain better tracking results, the parallelism of quantum computing will be introduced to the immune cloning algorithm among is used to solve the three-dimensional human motion tracking problems. Because quantum computing has the superposition and parallelism, therefore optimizing the high-dimensional functions have a great advantage, and has good search capabilities and search speed, suitable for this type of high-dimensional non-linear human tracking, and relatively high real-time requirements The high-dimensional optimization problem.On the basis of a survey of the present research work of computer vision-based 3D human motion tracking, this thesis studies the images feature subtraction, the reconstruction of human model, filtering algorithm and quantum optimization algorithm. The main contributions of this thesis are follows:1. Improved an automatic method for initialization of joint points and undetected points prediction. Aiming at the problem that setting the initial joint points manually in the proposed work, this thesis presents an automatic method for initialization of joint points and prediction of undetected points. These joint points lie on the skeleton line of a thinned region and parameters are estimated by minimizing the distance between the detected joint points and the projection of 3D model.2. Present a three-dimensional human motion tracking method based on Quantum Evolution algorithm. According to the updates and variation ideas of Evolutionary algorithm, designed the chromosomes and population that have been represented by probability. A high-dimensional optimization problem like the human tracking algorithm can be solved by changing the populations of the probability bits, and that can get the global optimal solution very good in the search space.3. Proposed a three-dimensional human motion tracking method based on Quantum-Inspired Immune Cloning Algorithm (QICA). Its main idea is to solve a single hypothetical optimization problem based on quantum superposition and parallelism, in which using the immune selective operation obtain the variation of the quantum populations, using cloned operation to extend the quantum populations, so the search space is enlarged further, and then it will obtain more favorable for the optimal solution.