Figure 3
From: Automatic Reasoning about Causal Events in Surveillance Video
The exemplar, or training, database is constructed for the descriptor feature set (both gaze direction and spatiotemporal action) and a PCA binary tree constructed. This image shows one level of the binary tree with the indices into frames on the y-axis and the node on the x-axis. A shaded block represents the occupancy of that frame at that node. The nodes shown are at depth 3 of the tree. This demonstrates that the tree is quite evenly split, which is important for traversal when searching.