Research On Multi-feature Fusion Recognition Of Incorrect Sit Posture

With the development of society, our children academic pressure is growing and learning time is more and more.The incidence of myopia also become higher and higher. Expert analysis showed that prolonged bad writing posture is one of the main causes of myopia. Paper books can not be achieved in the supervision of students writing posture. Children mainly are reminded by teachers and parents and worn ear-hook or spinal braces to correct sitting posture after a problem with the physical condition. The writing teaching system provides the technical condition for the real-time supervision of writing sitting posture. As an important research branch of the interactive structure of the system, the writing sitting posture detection is an important, which provides a real and reliable sitting position data for Automatic reminder and correction of the user's bad sitting posture.At present, the sitting posture recognition is mainly based on the single class feature, such as the geometric feature method based on the user profile, which is a kind of harmful human body posture alarm scheme, and the sitting recognition method based on the skin color feature.The main disadvantage of single feature recognition is that non-standard writing posture recognition rate is low. According to the young user's some unique non-standard writing posture,we proposed multiple feature fusion recognition method. The main work is as follows:First, the correlation detection recognition algorithms involved in the posture detection process are described in detail,such as moving object detection, skin color feature extraction, SURF feature extraction and multi classification algorithm and other algorithms. Their advantages and disadvantages are compared.which provides theoretical support for the follow-up to this article.Second, the principle of sitting posture is analyzed by a physiological point.According to children's sitting characteristics, we classify incorrect sitting posture for writing induction into seven groups, such as lying to write, humpback and so on.Firstly, considering the clustered skin area in a fixed region of YCbCr space which had ellipse-like projection in CbCr plane, extracted the skin area of motion object under the different brightness;Then extracted SURF feature using different thresholds.Third, Because single feature recognition rate of incorrect writing sitting posture is low, the way that using the multi feature fusion to identify incorrect sitting posture were proposed. Calculated the same position of the different features of the fusionweight by single class feature classification to get the correct rate of all kinds of sitting posture recognition, that is to say, the sum of the weight being positive proportional relationship with recognition rate of the same kind of position of the different characteristics is one. Then weighted fusion of different features in similar posture and maked use of BP artificial neural network to classify.Fourth, we implement non-standard sitting posture monitoring program. Based on the way of weighted fusion of different features in similar posture, we designed a simulation system to test and verify the feasibility of the method.The experimental results show that the recognition rate of this method is better than that of single class feature method, and it has better practicability. Which provides a real and reliable sitting position data for Automatic reminder and correction of the user's bad sitting posture.