A new multisensor software architecture for movement detection: Preliminary study with people with cerebral palsy

Abstract

A five-layered software architecture translating movements into mouse clicks has been developed and tested on an Arduino platform with two different sensors: accelerometer and flex sensor. The archi-tecture comprises low-pass and derivative filters, an unsupervised classifier that adapts continuously to the strength of the user's movements and a finite state machine which sets up a timer to prevent in-voluntary movements from triggering false positives. Four people without disabilities and four people with cerebral palsy (CP) took part in the experi-ments. People without disabilities obtained an average of 100% and 99.3% in precision and true positive rate (TPR) respectively and there were no statistically significant differences among type of sensors and placement. In the same experiment, people with disabilities obtained 97.9% and 100% in precision and TPR respectively. However, these results worsened when subjects used the system to access a commu-nication board, 89.6% and 94.8% respectively. With their usual method of access-an adapted switch- they obtained a precision and TPR of 86.7% and 97.8% respectively. For 3-outof- 4 participants with disabilities our system detected the movement faster than the switch. For subjects with CP, the accelerometer was the easiest to use because it is more sensitive to gross motor motion than the flex sensor which requires more complex movements. A final survey showed that 3-out-of-4 participants with disabilities would prefer to use this new technology instead of their tra-ditional method of access.