Methods for person identification on a pressure-sensitive floor: Experiments with multiple classifiers and reject option

Analysis of spatial domain information for footstep recognition

“…Experiments show that reduction of the sensor density affects significantly the recognition performance. This indicates that a relative high density is necessary and might well contribute to the good recognition performance reported here compared with that in related publications [2], [9].…”

“…The first was the popular ground reaction force (GRF) used previously in [2], [4], [8], [11], [15], in this case an average across all sensors was carried out to obtain a global profile for the GRF. The other two features were the spatial average of the sensors, which results in a single average profile of all sensors of the footstep signal; and finally the upper and lower contour profiles of the time domain signal.…”

“…Data dimensionality is also reduced using principal component analysis (PCA) [16], retaining more than 96% of the original information by using the first 140 principal components. as the kernel, due to very good performance in previous studies in this area [2], [3].…”

“…On the other hand, different types of sensors that capture transient pressure [4], [8], [9], [10], [11] have been used with relatively low sensor density (typically 9 sensors per m 2 for the given results respectively). A similar analysis was presented in [12], but focusing on the temporal information of the signals.…”

Analysis of Time Domain Information for Footstep Recognition

“…The identi cation rates achieved of around 80-90% are promising and give an idea of the potential of footsteps as a biometric [7,8]. However, these results are related to relatively small databases in terms of number of persons and footstep signals, typically around 15 people and perhaps 20 footsteps per person [5]; this is a limitation of the work to date.…”