The Hopkins155 dataset consists of 156 video sequences containing two or three actions. The motion of each video sequence corresponds to a low-dimensional subspace. Each video sequence has 39-550 data vectors extracted from two or three motions.

Sequences can be roughly divided into three categories:

• Checkerboard Sequences: This category includes 104 sequences of indoor scenes shot with a handheld camera under controlled conditions. A checkerboard pattern on the objects is used to ensure a large number of tracking points.

• Traffic Sequences: This category consists of 38 sequences of outdoor traffic scenes shot with a moving handheld camera.

• Other (Articulated/Non-rigid Sequences): This category contains 13 sequences showing motion constrained by joints, head and face motion, people walking around, etc.

Example data: