0

I am working on action recognition topic using Matlab. Each video contains specific number of frames, and each frame contains different coordinates of X,y,z.

After extracting the features, I got an cell array which contains 3D array for each video. Lets be clear with numbers: I have 40 videos, which give me 40x1 cell array. Each cell contains 128x3xN where N is number of frames. have corresponding labels array 40x1 double array.

Example of the first 10 videos:

<128x3x54 double>
<128x3x43 double>
<128x3x43 double>
<128x3x55 double>
<128x3x31 double>
<128x3x45 double>
<128x3x45 double>
<128x3x33 double>
<128x3x38 double>
<128x3x38 double>

I want to classify these videos using libSVM, how I can change the dimensions to 2D array and preserve the meaning of features ? And having similar aray dimensions for each video ?

matlab user
  • 79
  • 8
  • do you want each individual video to be represented by a 2D array? So that you'd get a M x N array, where M is the length of the frame information vector and N is the number of frames? Also, to get similar array dimensions for each video, you would have to trim down your videos to the length of the shortest video, is that what you mean? – Wouter Kuijsters Mar 18 '15 at 10:19
  • to apply SVM on the data, I want AxB data array and Ax1 labels array; where A is number of samples, and B is number of features. I have here A=40, and B is 128; but my problem is I have many frames, so I won't have similar B for all videos. I am not sure about triming, it would give good results ? – matlab user Mar 18 '15 at 11:53

1 Answers1

0

for example

10 action * 40 videos for each * (128x3xN) features = input for SVM for training which create 400 rows with (128x3xN) features and need 400 labels for training

for the testing you have 1 action * 1 video * (128x3xN) features in order to predict the action from trained SVM

Video Analysis Deep Learning
  • 338
  • 2
  • 12