The characterization of daily activities using accelerometers is a currently active research field, with special interest on fall detection of elderly people and sports performance. Most works that account walking and running are peak-acceleration based, however, false positives due to artefact acceleration peaks affect the estimation. Also, the proposed algorithms must be simple enough to be implemented in a smartphone or an embedded device. In this work a non-peak based methodology is proposed and validated with three publicly available databases. As a result, the methodology provides a quality-of-activity estimation, useful for sports performance and avoiding false positives in fall detectors.