Human Activity Recognition Using Multidimensional Indexing

“Human activity recognition from a sequence of angular poses and velocities of the main human body parts.” • An activity is represented by a set of pose and velocity vectors for the major body parts (hands, legs, and torso) and stored in a set of multidimensional hash tables. • Each body part has a separate hash table which includes all the model activities. • Recognize the activity invariant to the activity speed/time shift. • It is claimed that it is robust to partial occlusion since each body part is indexed separately. • It also has a view angle robustness of ± 30 degrees. The 9 body parts & feature vectors 1 1 ( , ) θ θ • The 9 body parts: i. Torso (and head) ii. Upper left arm iii. Upper right arm iv. Lower left arm (forearm + hand) v. Lower right arm vi. Upper left leg (thigh) vii. Upper right leg viii. Lower left leg (calf + foot) ix. Lower right leg • 18-dimensional feature vector • :angle with x-axis • :angular velocity • The angular velocities are calculated as the difference of angular positions of two successive frames. 2 2 ( , ) θ θ