Toward the Unaided Estimation of Human Walking Speed Based on Sparse Modeling

The ability to track people's movement (e.g., the human walking) is an important topic of interest in numerous research areas. State-of-the-art systems based on, e.g., dead reckoning, usually do not need any additional infrastructure. The disadvantage of the method is the poor accuracy because of measurement noises. The typical approach to noise cancellation is zero-velocity update (ZUPT). The method takes advantages of the cyclical nature of human walking. Thus some limitations of the ZUPT approach we propose the novel method based on transient artifact reduction algorithm (TARA) to mitigate these errors by noise estimation. TARA is the novel approach to estimate the low-frequency components in the signals by applying the concept of sparse optimization. The proposed algorithm was validated on synthetic and real signals. The tests for synthetic signals were performed for pulse signals with noise characterized by the Hurst exponent. In real signals studies, we applied the signals from three different data sets that were prepared to test the algorithm to human walking estimation. We found that TARA can estimate the underlying noise in the dead-reckoning-based systems. The results show that the TARA method can be used to improve the accuracy of human walking speed estimation as an alternative to the state-of-the-art method such as ZUPT. The results prove to be the important step toward unaided estimation of human walking velocity based on dead-reckoning technique.