Research Of Gait Pattern Classification Based On Accelerometer

In recent years, as the improvement of living standard, changes of lifestyle and livingenvironment, people’s physique has been changed. Monitoring of exercise energyconsumption is significant for people’s plan of scientific exercise to benefit one’s physique.Pedometer based on acceleration sensor is helpful to monitoring of exercise energyconsumption. However, almost all of the studies were based on walking and running, otherthan upstair and downstair. Studies have shown that the pedometer underestimated the energyconsumption of the up and down stairs. The classification of gait pattern is necessary tomonitoring of energy consumption of going upstair and downstair.Three gait patterns (walking, up stair and down stair) based on accelerometer werestudied in the paper.1) A device of picking up acceleration signal was designed. Tri-axisaccelerometer and controller STM8L152C were used to collect the signals. The device wasplaced on the man’s upper arm which worked when walking, upstair and downstair.2) Analgorithm of gait pattern classification was studied. First, every signal was separated into threedirections: X axis, Y axis, Z axis. The signals of acceleration sum were calculated andsegmented with sliding window. Then, an algorithm of sample entropy and wavelet energyabout extracting gait pattern was proposed, and then the decision tree classifier and Bayesclassifier were used for classification. The results showed that the device might extractacceleration signals in upper arm effectively. The general classification accuracy of decisiontree and Bayes classifier were up to75%and78.75%which improved15.85%and19.17%compared with the wavelet energy alone.Three gait pattern classification of walking, upstair and downstair was realized with thedevice and its algorithm. It indicates that the method based on sample entropy and waveletenergy is superior to the wavelet energy alone, and have an applicable value.