Lost frame Detection and Recovery Model in a Multiview Video Transmission over Wireless Sensor Network using Neural Network

Abstract

Multiview Video (MVV) is one of the emerging technology in recent years. The concept of MVV is becoming very important, during the implementation of 3D systems to enhancing the viewing of high-resolution videos and images from dif ferent angles. Streaming of MVV over a wireless sensor network (WSN) is very susceptible to whole-frame losses due to wireless channel errors and low-resolution cameras are used as sensor nodes in WSN. Along this, different studies try to develop error concealment techniques for MVC. In this thesis, we propose the identification of a lost frame method by using a Machine Learning (ML) models and a recovery algorithm for a lost frames of MVV in WSN by using Long Short Term Memory (LSTM) regression method. The detection method uses video and image quality assessment techniques to extract the features from the MVV frame sequences. The recovery method uses motion estimation and disparity estimation techniques to extract and select features for LSTM regression algorithm from MVV frame sequences. The performance of the proposed methods was analyzed on different MVV se quences. The experimental results of the proposed detection method have scored 93.12 % accuracy to detect the lost frames in MVV sequences. And the proposed LSTM based recovery algorithm has the capability to improve the video quality of MVC at the decoder side. Compared with the recent methods the proposed method exceeded the average Peak Signal to Noise Ratio (PSNR) upto 2.47dB. The complexity of the proposed method also acceptable. This study is expected to open up new perspectives on how to detect and restore the missing frames in MVV transmission at real-time