Gait Recognition - Hu Moments Classification
Once the Hu moments have been generated, the mean and variance for each of them is determined. This results in six values which can be used for classification. These results are normalised and then compared with the normalised training set. The top five matches are determined using the shortest Euclidean distance technique, and ranked accordingly. If there is a majority vote among the ranking then the match is found. If there is no majority then the match with that largest vote and the overall highest ranking is chosen. Only the test images are classified on-the-fil. Each sub directory of images has a csv file with the pre-computed hu values.
The applet allows two different gaits can be loaded and viewed. The left hand gait is the test gait and the right hand gait is a gait selected from the training set. The arrow keys "<" and ">" allow you to move back and forth through the gait cycles and the "<<" and ">>" allow you to move to the first and last images of the gait cycles. The "Play" button can be used to view a simultaneous animation of the two gaits. The speed of this animation can be changed by modifying the value in the "Speed" combo box.
Once you have selected your test gait you can calculate the Hu moments for it by clicking on "Get Hu Moments". This will display the mean and variance of the Hu1, Hu2 and Hu8 moments.
Once you have generated the Hu moments you can classify them. By clicking on "Classify Data", the top five matches for each moment and the combined moments are displayed, along with the Euclidean distance between the test gait and matched training gait, in the following format:
SubjectNo : GaitNo (EuclideanDistance)
The final row of the table displays the 5-NN match subject for each of the three moments and the combined moments.
The Java Applet is shown below:
Download this applet:
zip file coming soon...