The Application Of Random Forest Algorithm In Body Posture Recognition Research

With the constantly improvement of our living standard and the rapid development of intelligent recognition technology, body posture recognition has become a hot topic in the field of computer vision, which has great application prospects. Nowadays we have already applied it to all aspects of life, such as the behavior analysis, human-computer interaction,smart home and so on. Functionally similar to body language, body posture is the natural language which can also convey information and send instructions. In recent years, a large number of algorithms have been used in body posture recognition. All of them are based on template matching, state space, or semantic description. However, these algorithms have some defects in different degrees, such as the low detection accuracy, bad stability, low system efficiency, higher requirements on hardware and so on. In order to solve the above problems, we hope to combine the sensor technology and computer communication technology to design a system that can accurately identify the body posture. Firstly, the system can get the changing path gain data of the body posture from the sensor fixed in all parts of the body. Then the data is transmitted to the computer control terminal by the established wireless body area network between the body and the sensor network. Finally,the random forest algorithm is introduced to process the data and recognize the body posture.Random forest algorithm is built on the basis of multiple decision tree model, and is widely used in the classification. As an excellent combination algorithm, it also provides a new way for body posture recognition. The data can be trained to get the learner model, then posture recognition is performed. The predicted results of body posture show that the algorithm can improve the accuracy of posture recognition. The relational graph of the out of bag error and the number of decision tree is obtained, which shows the out of bag error decreases in the early time with the increase of the number of decision tree, and then tends to be stable. From the order of the importance of feature variables in anechoic chamber and laboratory, we can clearly see that the brain has the greatest influence, which is consistent with the physiological characteristics of the human body and the structural function, and provides a theoretical basis for selecting feature variables. The relational graph between the out of bag error and the number of characteristic variables determines the number of selecting feature variables. The setting of these parameters provides a theoretical basis for the optimal model of random forest algorithm.As a kind of ensemble learning algorithm, the random forest algorithm is applied to the body posture recognition, which is a relatively new field of application. The results of the experiment show that the random forest algorithm, which has good robustness to noise and can prevent overfitting, significantly improves the accuracy and reliability of the body posture recognition and provides a new kind of method and idea for the wearable sensor equipment and the unattended monitoring.