Isolated Word Real Time Ethiopian Sign Language Recognition for Sexual Reproductive Health Using Kinect

There above 1.15 Million peoples with deafness in Ethiopia with their primary means of
communication called EthSL which is one of the linguistically minor languages in Ethiopia.
Automatic Sign Language Recognition is currently in its infant stage while automatic speech
recognition is commercially available nowadays. Currently all commercial translation services are
human based, and therefore expensive, due to the need for experienced translators. There are very
few sign language interpreters in Ethiopia in hospitals and interpretation prices and very costly, in
addition to the cost of interpreters there is a series problem related to Sexual Health Reproduction
where the patient personal information is exposed to the interpreter or family member and friend
most of the time so the patient will not tell some of the required information to the doctor due to
the third party interference, this research work is dedicated to overcome this problem which
proposes a real-time isolated word sign language recognition for selected SRH words in hospitals
and different health centers. This thesis used a depth sensing device called Kinect to capture the
sign language sequences and pass it to our classification algorithms.
The performances of the selected algorithms for online as well as offline modes of operation have
been tested and results show that there is a high level of decrease in accuracies for online mode of
operation compared to the results obtained on offline mode of operation. Among the five
experiments that have been carried out the first experiment using 1NN-DTW in online signerdependent mode outperforms the results of the other four experiments which achieved a total
accuracy of 80% without using skeleton smoothing techniques and 93.75 using skeleton
smoothing. The overall experiments result shows that real time systems are more susceptible to
environmental noise and prone to small signing variations.