Gait Learning-Based Regenerative Model: A Level Set Approach

Abstract

We propose a learning method for gait synthesis from a sequence of shapes(frames) with the ability to extrapolate to novel data. It involves the application of PCA, first to reduce the data dimensionality to certain features, and second to model corresponding features derived from the training gait cycles as a Gaussian distribution. This approach transforms a non Gaussian shape deformation problem into a Gaussian one by considering features of entire gait cycles as vectors in a Gaussian space. We show that these features which we formulate as continuous functions can be modeled by PCA. We also use this model to in-between (generate intermediate unknown) shapes in the training cycle. Furthermore, this paper demonstrates that the derived features can be used in the identification of pedestrians.

Robust gait recognition from extremely low frame-rate videos

“…Our method has 3 desirable properties: 1) the performance is less sensitive to the feature number used in the base classifiers; 2) it yields satisfactory results for extremely low frame-rates gait sequences under the influences of large/small gait fluctuations; 3) the performance can be further enhanced if the gallery videos have higher frame-rates. Compared with the temporal reconstruction-based methods [6,8,9], our method delivers significant improvements in terms of performance or generalization.…”

“…3. Although ERTSR outperforms our method on DB-high dataset, it should be pointed out that ERTSR assumes [6] 52 N/A PTSR [8] 44 N/A ERTSR [9] 87 N/A AG+PCA+NN 69 67 AG+NN 68 68 AG+RSM (our method) 80.40±1.35 80.60±1.26 the same motion among the gait periods, which is not applicable when there are large gait fluctuations (e.g., DB-low dataset) [9]. 4.…”

“…Based on 1fps gait videos, we compare our method with several temporal reconstruction-based, i.e., morphing-based reconstruction (Morph) [6], Periodic Temporal SR (PTSR) [8], and Example-based and Reconstruction-based Temporal SR (ERTSR) [9]. Due to the random nature of our method, we repeat each experiment 10 times and report the mean rank-1 CCRs (with the standard deviation) in Table 1.…”

“…A unified framework of example-based and reconstruction-based periodic temporal SR was proposed in [9], and it works well for gait recognition from (dual) extremely low frame-rate videos (with small gait fluctuations). Most of these works attempt to recover the high frame-rate gait sequences in the first place, yet in the applications of human identification, they either face high levels of reconstruction artifacts [6,8] for extremely low frame-rate input videos (e.g., 1fps), or assume the same motion among the gait periods, which is not applicable when there are large gait fluctuations [9].…”

“…Several temporal reconstruction-based methods were proposed to deal with such dual low frame-rate problems. In [6], Al-Huseiny et al proposed a level-set morphing approach for temporal interpolation. In [7], temporal Super Resolution (SR) method was employed to construct a period of high frame-rate gait sequence based on multiple periods of low frame-rate gait sequences.…”