Rule-based activity recognition

Abstract

Unlike the learning-based approach, where one does not need to know the details of a rehabilitation exercise, the rule-based approach requires a detailed description of each exercise to be recognized and evaluated in terms of a set of rules. Exactly because of this, it is difficult for the learning-based approach to provide specific feedback to patients regarding exactly what was performed correctly and what requires improvements. The rule-based approach could provide much more specific feedback to the patients. Studies on rule-based activity recognition can be roughly divided into two categories: (1) ad hoc rules that are applicable to a particular rehabilitation exercise; and (2) general purpose rules that are designed to define a broader range of exercises or activities. Another perspective related to rehabilitation exercises is the tolerance on movements within which are considered correct.For both types studies, such uncertainty can be captured explicitly by a threshold in the measurement as part of the rule, or by using the fuzzy interference using some membership definition. Fuzzy inference has also been used to detect activities.