Invariant trajectory indexing for real time 3D motion recognition

Abstract

Cochlear implants have become a standard treatment for many with severe to profound sensorineural hearing loss. However, delicate cochlear structures can be damaged during surgical insertion, which can lead to loss of residual hearing and decreased implant effectiveness. We propose a magnetic guidance concept in which a magnetically tipped cochlear implant is guided as it is inserted into the cochlea. In a scaled in vitro experimental study, we record insertion forces for nonguided and magnetically guided insertion experiments and compare the results. Results indicate that magnetic guidance reduced insertion forces by approximately 50%.

Parsing 3D motion trajectory for gesture recognition

“…16. The figure shows that as the ratio of trajectory decreases, the accuracy of Signature [32], IGS [44] and our method drop relatively small, due to the locality of our representation. The accuracy of the IGS descriptor is a little lower than that of the Signature descriptor because the global information used in IGS is not accurate.…”

“…Method Accuracy (%) CSS + HMM [18] 59.6 CDF + HMM [18] 64.1 Fourier [20] 71.8 Signature [32] 86.4 IGS [44] 91.3 Integral Invariants [35] 91.9 Our Method 93.4 MHAD datasets respectively. The ratios mean that only part of the whole trajectory is received up to the present frame and recognized online.…”

“…The proposed approach achieves a comparable accuracy, which is a good performance on the HDM05 dataset. The IGS [44] approach also has a high accuracy because this descriptor includes the global motion information of trajectories. However, the accuracy of the IGS approach will be much lower if only the primitive descriptor is used without the global information, and the global information is not robust for occlusion.…”

Flexible Trajectory Indexing for 3D Motion Recognition

“…The previous dynamic indexing method [10] is a primitive-level descriptor, which only remains the salient feature of primitive and filters the detailed information of point level descriptor. This makes motion recognition fast, but not flexible thus may not be able to handle complex motions.…”

“…Inspired by the dynamic indexing method [10], we use sub-primitive to represent 3D trajectory, and the signature descriptor [11] is used for point level description. Different from the primitive-level descriptor [10], the new approach represents trajectory in the sub-primitive level. Two trajectories represented by sub-primitives in comparison with primitives are shown in Figure 1.…”