Human Walking Behavior Recognition Technology Research On Kinect Based On Unity

All walks of life are contributing their wisdom and strength to build a safe society,which has led to the rapid technological development of mechanical vision and image recognition.In public places,a good intelligent monitoring system can prevent crimes from happening and help the police to deal with crimes.At the factory,robots with image processing functions can detect product indicators,which improve productivity and product quality.In entertainment places,"virtual reality" and "augmented reality" can educate and entertain people to feel the changes in life caused by technological advancement.Because of low research,development cost,flexible control and simple hardware,a perfect vision system has its certain market demand and scientific research value.To identify the walking behavior of the human body and improve the accuracy of previous algorithm recognition,this paper combines the Kinect camera with the Unity compiler,using the OpenCV function library,dynamic time wrapping algorithm,naive Bayes classifier and other imaging knowledge.The paper proposes the concept of dividing the motion area.After Setting up motion track templates,through machine learning,it can increase the recognition rate,and finally make an effective distinction.It can be applied in the field of intelligent monitoring,such as gait analysis,human body recognition,etc.In the field of image recognition,it also has certain creative work.In the design of image processing algorithm,this paper proposes an improved Canny algorithm.This algorithm replaces the original Gaussian filter with a bilateral filter,and replaces the original Sobel filter with a Scharr filter.Experiments show that the improved algorithm has better effect in the depth image of the human body.In the design of machine learning algorithm,this paper chooses the average template training method based on DTW algorithm,builds the DTW-NBC model with NBC classifier,then describes its training process in detail.This paper makes a certain comparative study on the selection of different models.Then it selects a vectorized direction angle variation sequence combined with the Unity screen division number as the motion model.Finally,after discussing the system design in detail,it is verified by contrast experiments that the human walking behavior recognition system can accomplish the target work.It provides theoretical and practical basis for the follow-up work.