Abstract
An IoT architecture can include real time analytics and historical analytics that run on device or other powerful servers. Hence streams can provide platforms for real time analytics within an IoT architecture. Even though streams in IoT essentially mean data streams; as they tend to grow bigger they would accommodate multimedia streaming. With multimedia and sensors, synchronization among the streams would play a vital part in many applications.
In this project we aim to achieve video synchronization to enhance indoor localization of objects. Consider a scene where objects change their locations frequently. We have an Opencv application with stereo module that assigns 3D coordinates to the objects. For the system to work in real time, the streams from different cameras (different view points of the scene) should arrive at the same time.
We use a multi media framework called Gstreamer and research on various methods to synchronize a stream efficiently up to < 1 frame sync. We tried experimenting with three different methods and then discuss which method ideally suits our application.
In this project we aim to achieve video synchronization to enhance indoor localization of objects. Consider a scene where objects change their locations frequently. We have an Opencv application with stereo module that assigns 3D coordinates to the objects. For the system to work in real time, the streams from different cameras (different view points of the scene) should arrive at the same time.
We use a multi media framework called Gstreamer and research on various methods to synchronize a stream efficiently up to < 1 frame sync. We tried experimenting with three different methods and then discuss which method ideally suits our application.
- Putting a buffer that can store 1 or 2 seconds worth of data on the receiver side.
- Using a common NTP clock for the devices belonging to the system and map the timestamps of packets to that clock.
- Using a common internal pipeline clock for sender and receiver.