Method for identifying behavior based on feature enhancement and decision fusion

Abstract

The invention discloses a method for identifying a behavior based on feature enhancement and decision fusion. The method comprises the following steps: pre-treating data of an accelerometer; calculating multi-dimensional feature vectors of the pre-treated data; using the forward sequence selection algorithm to select the optimal feature vector among the multi-dimensional feature vectors; using the Relief-F algorithm to select a proper feature value for feature enhancement; training one basic classifier and three weak classifier; taking the identification results from the basic classifier and the three weak classifiers as 4 parameters, meanwhile taking the identification result of the last action as another parameter, determining the weights of the 5 parameters through training; identifying human body behaviors by using secondary classification, a first classification using the basic classifier for identification, a second classification using weighted voting decision fusion to obtain a final identification result. According to the invention, the method can effectively differentiate similar actions that are difficult to differentiate, such as walking, going upstairs and going downstairs, and increase identification of human body behaviors.

Description

technical field [0001] The invention relates to the fields of a three-axis acceleration sensor, pattern recognition and the like, in particular to the field of human body behavior recognition based on three-axis acceleration data. Background technique [0002] Human behavior recognition using sensors has always been a research hotspot in the fields of sensor data processing and pattern recognition. Detecting people's daily behavior through a three-axis acceleration sensor has the characteristics of convenience and high recognition rate. However, acceleration data often has situations where it is difficult to distinguish multiple motion data, especially the three actions of walking, going upstairs, and going downstairs are very similar, and the difference between the data is very small even at a high-precision acquisition frequency. Once the data set becomes larger, there will be more mutual interference data between the three actions, and since the data collection platform ...

AI Technical Summary

Problems solved by technology

However, acceleration data often has situations where multiple motion data are difficult to distinguish, especially the three actions of walking, going upstairs, and going downstairs are very similar, and the difference between the data is very small even at a high-precision acquisition frequency

Once the data set becomes larger, there will be more mutual interference data between the three actions, and since the data collection platform is not necessarily fixed on the human body, the collected data will inevitably have errors due to jitter. In extreme cases , the data of these three types of motions may be indistinguishable, which makes it very difficult to correctly identify these three types of human motions

Images