Human Identification at a Distance
UMD's Evaluation Results on UMD Data

 
Home| Overview | People | Research | Publications | Data | Evaluation | Evaluation | Links | Contacts

The aim of our experiments conducted  was primarily to test the performance of our gait recognition methodology on video data captured by our outdoor surveillance cameras mounted at an oblique angle. We are investigating both frontal and fronto-parallel views for gait recognition.

Note: For all our experiments, the training and gallery sets are the same.
 
 
 
UMD Dataset 1

Image Labeling

MMDDYY_XXX_YYZZ where 
MMDDYY = date of video capture, 
XXX = subject id, 
YY = pl / pr / fa (walking direction : left to right, right to left, frontal) 
ZZ = 01 /02 (sequence number). 
 
 
 

Experiment 1:
How does the frontal gait recognition algorithm perform across the same view (frontal) of subjects captured on the same day? 
 
 
  Gallery
  Probe
Result Similarity matrix (image) Similarity matrix (matrix)
Similarity files


UMD Dataset 2
Image Labeling

The images are labeled as XXX_Y_ZZ where 
XXX = subject id, 
Y = 6/8 (camera id), 
ZZ = 01 / 02 (sequence number) 

Experiment 1:
How does the gait recognition algorithm perform across the same (fronto-parallel) view of subjects captured on different days? 
  
Gallery
  Probe
Result (HMM Approach) Similarity matrix (image) Similarity matrix (text)
Similarity files CMS Curve


 


Site First Created: 08/03/01  |  Last Modified: 04/04/02