Claims:WE CLAIM:
1. method for detecting pose of a subject in real-time, the method comprising:
identifying, by a pose detection system (103), a plurality of nodal points (211) corresponding to the subject from an input image frame (102) of the subject;
identifying, by the pose detection system (103), a skeleton pose (213) of the subject based on the plurality of nodal points (211);
computing, by the pose detection system (103), a feature descriptor for the skeleton pose (213) based on the plurality of nodal points (211);
comparing, by the pose detection system (103), the feature descriptor of the skeleton pose (213) with a plurality of predetermined feature descriptors (107) using a pose detection model (105),
wherein the pose detection model (105) is pre-trained with each of the plurality of predetermined feature descriptors (107), corresponding to a predefined pose of the subject, using predetermined deep learning techniques,
wherein each of the plurality of predetermined feature descriptors (107) are computed based on skeleton key points obtained by rotating a plurality of nodal points (211), corresponding to a reference image frame (102) of the subject, in a three-dimensional vector space; and
detecting, by the pose detection system (103), the pose of the subject as the predefined pose corresponding to one of the plurality of predetermined feature descriptors (107) based on the comparison.

2. The method as claimed in claim 1, wherein the skeleton pose (213) of the subject is identified by joining the plurality of nodal points (211) based on one or more nodal arrangements retrieved from predefined node detection libraries.

3. The method as claimed in claim 1, wherein computing the feature descriptor for the skeleton pose (213) comprises:
identifying a subset of the plurality of nodal points (211) based on spatial difference between each of the plurality of nodal points (211);
generating a feature vector corresponding to each pair of the plurality of nodal points (211) in the subset of the plurality of nodal points (211);
computing distance values between each of the feature vectors; and
ordering the distance values in a predefined sequence for computing the feature descriptor.

4. The method as claimed in claim 1, wherein rotating the plurality of nodal points (211) comprises rotating the plurality of nodal points (211) along a predetermined angle with respect to the reference image frame (102) of the subject.

5. The method as claimed in claim 1, wherein detecting the pose of the subject comprises:
assigning a similarity score for each of the plurality of predetermined feature descriptors (107) based on comparison; and
detecting the predefined pose corresponding to one of a predetermined feature descriptor, among the plurality of predetermined feature descriptors (107), having highest similarity score as the pose of the subject.

6. A pose detection system (103) for detecting pose of a subject in real-time, the pose detection system (103) comprising:
a processor (203); and
a memory (205), communicatively coupled to the processor (203), wherein the memory (205) stores processor-executable instructions, which on execution, cause the processor (203) to:
identify a plurality of nodal points (211) corresponding to the subject from an input image frame (102) of the subject;
identify a skeleton pose (213) of the subject based on the plurality of nodal points (211);
compute a feature descriptor for the skeleton pose (213) based on the plurality of nodal points (211);
compare the feature descriptor of the skeleton pose (213) with a plurality of predetermined feature descriptors (107) using a pose detection model (105),
wherein the pose detection model (105) is pre-trained with each of the plurality of predetermined feature descriptors (107), corresponding to a predefined pose of the subject, using predetermined deep learning techniques,
wherein each of the plurality of predetermined feature descriptors (107) are computed based on skeleton key points obtained by rotating a plurality of nodal points (211), corresponding to a reference image frame (102) of the subject, in a three-dimensional vector space; and
detect the pose of the subject as the predefined pose corresponding to one of the plurality of predetermined feature descriptors (107) based on the comparison.

7. The pose detection system (103) as claimed in claim 6, wherein the processor (203) identifies the skeleton pose (213) of the subject by joining the plurality of nodal points (211) based on one or more nodal arrangements retrieved from predefined node detection libraries.

8. The pose detection system (103) as claimed in claim 6, wherein to compute the feature descriptor for the skeleton pose (213), the processor (203) is configured to:
identify a subset of the plurality of nodal points (211) based on spatial difference between each of the plurality of nodal points (211);
generate a feature vector corresponding to each pair of the plurality of nodal points (211) in the subset of the plurality of nodal points (211);
compute distance values between each of the feature vectors; and
order the distance values in a predefined sequence for computing the feature descriptor.

9. The pose detection system (103) as claimed in claim 6, wherein the instructions cause the processor (203) to rotate the plurality of nodal points (211) along a predetermined angle with respect to the reference image frame (102) of the subject.

10. The pose detection system (103) as claimed in claim 6, wherein to detect the pose of the subject, the processor (203) is configured to:
assign a similarity score for each of the plurality of predetermined feature descriptors (107) based on comparison; and
detect the predefined pose corresponding to one of a predetermined feature descriptor, among the plurality of predetermined feature descriptors (107), having highest similarity score as the pose of the subject.