Virtual Human Visual Perception And Memory Model Based On CUDA And Its Application In Path Planning

Virtual reality technology is a cutting-edge cross-disciplinary technology depending on the development of graphics, stereoscopic display technology, sensor technology, artificial intelligence and pattern recognition technology, and it has made a great progress. Its application has been involved in the military, engineering, medicine, education, entertainment and many other fields. As an important branch of virtual reality technology, autonomous virtual human technology is a research hotspot in recent years, which aims to imitate the intelligent life's geometric features, motion characteristics, cognitive patterns and behavioral decision ability in the real world. Visual perception plays a crucial role in human perception channels. As is showed in studies, that 80% of human's environmental information is obtained through vision. Therefore, the research on virtual human's visual perception and memory model is significant.Based on the background mentioned above, this paper studies the parallel method of virtual human's visual perception and memory model, and applies this model to virtual human's path planning successfully. The following studies are carried out in this thesis:(1) Firstly, the traditional methods of virtual human's visual perception are studied, which include ray casting method based visual perception and image depth buffer information based visual perception. The former is a geometric method, which is simple and easy to understand. The latter is a method based on the principle of computer graphics, so it is relatively complex, but can simulate the human visual perception ability more realistic. Based on the previous research, depth buffer based visual perception theory is presented, which is adapted to the experiment requirement in this paper. Secondly, the memory model of virtual human's perception information is studied, including function modules such as memory processing, short-term memory and long-term memory, memory saving and forgetting. Finally, The definition of key point adapted to the model and the algorithm in this paper are given.(2) Serial method of virtual human's visual perception and memory model is studied. Besides, the data structure and algorithm flow are introduced. According to the serial method, a parallel implementation method based on CUDA parallel computing framework is proposed. This method can make fully use of the hardware performance of the heterogeneous computer system, and improve the computing speed of virtual human's perception and memory model. It has a broad application prospect for large complex virtual scene.(3) Virtual human's collision avoidance and path planning based on visual perception information is studied. A two-level global path planning algorithm is proposed, which can improve the computing efficiency. The algorithm avoids the efficiency problem which is caused by mesh dividing and can imitate human being's perception and memory ability effectively.The experiment is accomplished by a computer, which is equipped with a NVIDIA GTX 750 Ti in Windows7 OS. Blender Python interface is used to develop the program. The experimental results showed that the visual perception and memory model can be successfully applied to the path planning of virtual human, and provide technical support for improvement of the execution efficiency in traditional virtual human's path planning.