Detecting Indian Sign Language using a HandPose model. For now, it detects all the alphabets.
Most sign language detection models prefer using Skin Segmentation for preprocessing. However, there are a lot of limitations to this method, like lighting issues, different skin tones, and distinguishing hand skin from any other skin.
A better approach would be to use the MediaPipe model. The idea here is to extract the skeleton and build a CNN model with this data. Here, the code and model was adapted from LearnOpenCV's HandPose repo.
After training the model, it can detect any ISL single-handed alphabet, by first extracting its keypoints, and passing its skeleton through the CNN model.
There are lack of good quality ISL datasets available online. The lack of quality leads to the HandPose not being able to find any keypoints, hence training a poor model.
Again, due to a lack of quality ISL datasets - there aren't any Double-Handed ISL datasets freely available. Single-Handed ISL is very similar to American Sign Language (ASL) and won't be as practically useful as Double-Handed ISL is used more.
- Training model to recognize numbers
- Training model to recognize words (through Gesture Recognition)
- Applying the model to Livestream Data, making this a Live-Time ISL detection system