SVCL Motion Classification

Library Entrance ...Scene Description [html] [pdf]
Congestion Levels Experiment Oblique Angle				Experiment Accuracy
Congestion Levels Experiment Oblique Angle				Martin NN = 64% Nearest-neighbor classifier using Martin distance. State KL NN = 57% Nearest-neighbor classifier using state KL divergence State KL SVM = 72% SVM classifier using state KL kernel Image KL NN = 72% Nearest-neighbor classifier using image KL divergence Image KL SVM = 71% SVM classifier using image KL kernel
Description In this experiment, we discriminate between congestion/flow levels: high, medium, and low.	Clips: high divx mov mp4 divx 1 2 mov (h.264) 1 2 mp4 1 2	Clips: medium divx mov mp4 divx 1 2 3 4 5 6 7 8 mov (h.264) 1 2 3 4 5 6 7 8 mp4 1 2 3 4 5 6 7 8	Clips: low divx mov mp4 divx 1 2 3 mov (h.264) 1 2 3 mp4 1 2 3

Entering/ Exiting Experiment Oblique Angle				Experiment Accuracy
Entering/ Exiting Experiment Oblique Angle				Martin NN = 62% Nearest-neighbor classifier using Martin distance. State KL NN = 51% Nearest-neighbor classifier using state KL divergence State KL SVM = 44% SVM classifier using state KL kernel Image KL NN = 37% Nearest-neighbor classifier using image KL divergence Image KL SVM = 67% SVM classifier using image KL kernel
Description In this experiment we attempted to discriminate between pedestrian traffic exiting versus entering versus balanced. Location: UC San Diego's Geisel Library 4th floor facing south.	Clips: mostly entering divx mov mp4 divx 1 2 3 4 5 mov (.h264) 1 2 3 4 5 mp4 (mpeg4) 1 2 3 4 5	Clips: balanced divx mov mp4 divx 1 2 3 4 5 6 7 mov (h.264) 1 2 3 4 5 6 7 mp4 (mpeg4) 1 2 3 4 5 6 7	Clips: mostly exiting divx mov mp4 divx 1 2 3 4 5 mov (.h264) 1 2 3 4 5 mp4 (mpeg4) 1 2 3 4 5

Front Angle #1		Need other levels to discriminate.	Need other levels to discriminate.	No Experimental Results
Description In this experiment, we discriminate between congestion/flow levels: high, medium, and low. Video compressed into three formats: divx, mov (.h264), and mp4 (mpeg4). All at 30fps 320 x240	Clips: Angle #1 divx mov mp4 divx 1 2 mov (h.264) 1 2 mp4 (mpeg4) 1 2

Front Angle #2		Need other levels to discriminate.	Need other levels to discriminate.	No Experimental Results
Description In this experiment, we discriminate between congestion/flow levels: high, medium, and low. Video compressed into three formats: divx, mov (.h264), and mp4 (mpeg4). All at 30fps 320 x240	Clips: Angle #2 divx mov mp4 divx 1 mov (h264) 1 mp4 (mpeg4) 1

Side Angle Still		Need other levels to discriminate.	Need other levels to discriminate.	No Experimental Results
Description In this experiment, we discriminate between congestion/flow levels: high, medium, and low. Video compressed into three formats: divx, mov (.h264), and mp4 (mpeg4). All at 30fps 320 x240	Clips: Still Side divx mov mp4 divx 1 2 mov (h264) 1 2 mp4 (mpeg4) 1 2

Side Angle Pan		Need other levels to discriminate.	Need other levels to discriminate.	No Experimental Results
Description In this experiment, we discriminate between congestion/flow levels: high, medium, and low. Video compressed into three formats: divx, mov (.h264), and mp4 (mpeg4). All at 30fps 320 x240	Clips: Pan Side divx mov mp4 divx 1 2 mov (h.264) 1 2 mp4 (mpeg4) 1 2

Oblique Angle Pan		Need other levels to discriminate.	Need other levels to discriminate.	No Experimental Results
Description In this experiment, we discriminate between congestion/flow levels: high, medium, and low. Video compressed into three formats: divx, mov (.h264), and mp4 (mpeg4). All at 30fps 320 x240	Clips: Panning divx mov mp4 divx 1 2 mov (h.264) 1 2 mp4 (mpeg4) 1 2