Experiments on UofM Data

Experimental Question

From the University of Maryland's database, the performance of our gait recognition methodology on video data captured by outdoor surveillance cameras mounted at an oblique angle is evaluated. Their database contains both frontal and frontal-parallel views for gait recognition.

The University of Maryland's database contains two data sets.

Dataset-1 contains walking sequences of 25 subjects in 4 different poses.
Dataset-2 contains walking sequences of 55 subjects walking along a T-shape pathway.

For a complete description of their database see their description page .

Experiments and Results

We generate tables that are similarity comparison (by finding the L2norm) between gallery and probe views for the following experiments. The similarity matrixes are presented as text files and cmc curves. The headings of the columns refer to the gallery, and the row headings are the probe view(s). The file labels for the headings can be found on UofM's evaluation page. For out technique there is not a training set.

Dataset-1
1. Experiment 1: How does the gait recognition algorithm perform across the frontal-parallel view of subjects?
  - Gallery
  - Probe
  - txt: (d1,d2,d3,d4)
  - cmc: (d1,d2,d3,d4)
2. Experiment 2: How does the gait recognition algorithm perform across the frontal view of subjects?
  - Gallery
  - Probe
  - txt: (d1,d2,d3)
  - cmc: (d1,d2,d3)
Dataset-2
1. Experiment 1: How does the gait recognition algorithm perform across the save view of subjects?
  - Experiment 1.1:
    - Gallery
    - Probe
    - txt: (d1,d2,d3)
    - cmc: (d1,d2,d3)
  - Experiment 1.2:
    - Gallery
    - Probe
    - txt: (d1,d2,d3)
    - cmc: (d1,d2,d3)
2. Experiment 3: How does the gait recognition algorithm perform across different cameras?
  - Experiment 3.1:
    - Gallery
    - Probe
    - txt: (d1,d2,d3)
    - cmc: (d1,d2,d3)
  - Experiment 3.2:
    - Gallery
    - Probe
    - txt: (d1,d2,d3)
    - cmc: (d1,d2,d3)
3. Experiment 4: How does the gait recognition algorithm perform across different days?
  - Experiment 4.1:
    - Gallery
    - Probe
    - txt: (d1,d2,d3)
    - cmc: (d1,d2,d3)
  - Experiment 4.2:
    - Gallery
    - Probe
    - txt: (d1,d2,d3)
    - cmc: (d1,d2,d3)

Experiment Specifics

Our paradigm of measuring static body parameters does not require any specific training set that helps define a model that is then applied to the gallery and the probes. When performing matching across views, all subjects are used to compute "cross-condition mapping functions". That is, a single linear regression was done for each measured feature in our method that would map from one condition to the other. All subjects were used because of the limited number of subjects. Also, we were unable to using our depth compensation method because of a lack of calibration information.

MAIN PAGE