Body Action Recognition Based On Quantum Neural Network

At present,human body action recognition is one of the research directions in the field of computer vision.It is widely used in important fields such as human-computer interaction,action analysis and video surveillance,which is a promising research topic with a huge market demand.However,how to extract human body actions from complex environments and accurately identify them are two challenging problems.Quantum computing is an interdisciplinary subject which combines quantum mechanics and computer science.The inherent characteristics of quantum entanglement and unique computational advantages have attracted the attention of researchers all over the world.In this paper,quantum computing is introduced into the field of computer vision,which provides a new concept and idea to analyze and recognize human body actions from the microscopic world.The main contributions of this paper are as follows:1.A novel n-bit quantum full adder is proposed.The full adder designed in the existing literature has the problem of large quantum cost,meanwhile,the new quantum full adder adopts the carry look-ahead mode without carry input,and judges the carry of addition and positive and negative sign of subtraction with the highest overflow mark bit that does not participate in the calculation of high and low bit,which does not increase time delay of the circuit.Therefore,it has the characteristics of low quantum cost and simple structure,which is beneficial to reduce the construction cost and physical difficulty of the quantum full-scale device.2.The entire action image of each frame is stored in the quantum entangled state by using the NEQR quantum image storage model.The background difference process of quantum image is designed to extract the foreground action image in the static environment.Using the quantum ViBe algorithm to extract the foreground action image in the motion environment.The biggest difference of ViBe detection algorithm in quantum state is that N random NEQR quantum images are constructed as the background model sample library instead of the background sample library of a certain position.Then,a convolution method based on quantum oracle is proposed to extract the feature points of the whole quantum image by convolution operation.3.An improved quantum BP neural network model is proposed.The difference from the quantum BP neural network model in the existing literature is that the input of the input layer is represented by qubit 0,and the phase of the phase is changed by the quantum revolving gate and then aggregated to obtain the neuron output.The experimental results showed that the improved quantum BP neural network model has better recognition ability than the existing quantum BP neural network model.4.An embedded system is designed to simulate the evolution of quantum systems in body action recognition.The system uses S5P6818 chip,built-in high-performance 8-core Cortex-A53 processor,equipped with Ubuntu14.04 operating system and development software.The quantum algorithm simulation program for writing body action recognition in C++ is used to train and identify the action quantum image feature points in the Weizmann Dataset video library and the video library collected in this paper,which verifies the feasibility of the whole recognition scheme.Through experiments,the whole system is not only stable compared with other solutions,but the recognition rate reaches 100%.