Efficient Step Detection and Step Length Estimation Method Based on Smartphones

Pedestrian dead reckoning (PDR) based on smartphones plays an important role in indoor positioning, activity recognition, and health management. However, existing PDR methods generally assume that smartphones are carried in a fixed posture and do not fully consider the step detection (SD) errors during nonwalking states, which poses significant limitations in diverse Internet-of-Things (IoT) application scenarios. Given the variety of smartphone postures and gait patterns, as well as the need for robust SD and step length estimation (SLE) methods, this article proposes the multicondition constraints and instantaneous frequency-based SD and estimation (MCCIF-SDE) method. First, the SD method combines adaptive peak-valley detection thresholds with dynamic window periodic constraints and introduces gait similarity and continuity constraints. This effectively reduces step-counting errors caused by false walking, gait variations, and smartphone carrying mode switches. Second, the SLE method achieves higher estimation accuracy by extracting instantaneous frequency information from acceleration data. The model is general and robust across different gait patterns and smartphone carrying modes, without the need for machine learning-based parameter adjustment, greatly reducing system complexity. Experimental results show that the proposed method achieves an average step-counting accuracy of 96.5% and an average SLE error as low as 1.8% under various smartphone postures, gaits, and false walking modes. This method significantly improves the performance and applicability of the PDR system in diverse scenarios.