Motion recognition via a two-dimensional symbol having multiple ideograms contained therein
Methods of recognizing motions of an object in a video clip or an image sequence are disclosed. A plurality of frames are selected out of a video clip or an image sequence of interest. A text category is associated with each frame by applying an image classification technique with a trained deep-learning model for a set of categories containing various poses of an object within each frame. A “super-character” is formed by embedding respective text categories of the frames as corresponding ideograms in a 2-D symbol having multiple ideograms contained therein. Particular motion of the object is recognized by obtaining the meaning of the “super-character” with image classification of the 2-D symbol via a trained convolutional neural networks model for various motions of the object derived from specific sequential combinations of text categories. Ideograms may contain imagery data instead of text categories, e.g., detailed images or reduced-size images.
1. A method of recognizing motions of an object in a video clip or an image sequence comprising:
selecting a plurality of frames out of a video clip or an image sequence of interest;
obtaining a detailed image in each frame by cropping said each frame to a bounding box that contains an object of interest including the object's pose;
forming a super-character by embedding respective bounding boxes of the plurality of frames as corresponding ideograms in a two-dimensional (2-D) symbol having multiple ideograms contained therein and the super-character representing a meaning formed from a specific combination of the multiple ideograms; and
recognizing a particular motion of the object by obtaining the meaning of the super-character with image classification of the 2-D symbol via a trained convolutional neural networks model for various motions of the object derived from specific sequential combinations of respective detailed images.
2. The method of claim 1 , wherein the 2-D symbol being a matrix of N×N pixels of K-bit data and the matrix being divided into M×M sub-matrices with each of the sub-matrices containing (N/M)×(N/M) pixels, said each of the sub-matrices representing one ideogram, where K, N and M are positive integers, and N is a multiple of M.
3. The method of claim 2 , wherein K is 5, N is 224, M is 4. M×M is 16 and N/M is 56.
4. The method of claim 2 , wherein K is 5, N is 224, M is 8, M×M is 64 and N/M is 28.
5. The method of claim 1 , wherein the trained convolutional neural networks model comprises bi-valued 3×3 filter kernels in a Cellular Neural Networks or Cellular Nonlinear Networks (CNN) based integrated circuit.
6. The method of claim 5 , wherein the trained convolutional neural networks model is achieved with following operations:
(a) obtaining a convolutional neural networks model by training the convolutional neural networks model based on image classification of a labeled dataset, which contains a number of multi-layer 2-D symbols, the convolutional neural networks model including multiple ordered filter groups, each filter in the multiple ordered filter groups containing a standard 3×3 filter kernel;
(b) modifying the convolutional neural networks model by converting the respective standard 3×3 filter kernels to corresponding bi-valued 3×3 filter kernels of a currently-processed filter group in the multiple ordered filter groups based on a set of kernel conversion schemes;
(c) retraining the modified convolutional neural networks model until a desired convergence criterion is met; and
(d) repeating (b)-(c) for another filter group until all of the multiple ordered filter groups are converted to the bi-valued 3×3 filter kernels.