Action Recognition of Insects Using Spectral Clustering

We propose a technique to recognize actions of grasshoppers based on spectral clustering. We track the object in 3D and construct features using 3D object movement in segments of video which discriminate between different classes of actions. We sample from these feature vectors and compute the eigenvalues and eigenvectors of affinity or similarity matrix. Then, we perform K-means algorithm to cluster component from each of dominant eigenvectors of the affinity matrix. These dominant eigenvectors are embedding coordinate of video segments in our embedding space. We experimented with our method on a noisy track of one insect to validate our approach.