Crowd Intelligence Empowered Video Transmission in Ultra-low-bandwidth Constrained Circumstances.

The performance of video transmission depends on their ability to efficiently improve in jitter, latency and bit rate. However, emerging video transmission needs create new difficulties owing to the devices in ultra-low-bandwidth constrained circumstances, such as no man's land, underwater acoustic communication, etc. And the existing adaptive bit rate algorithms can not apply to the extreme circumstances well. In this work, a adaptive problem of video transmission is studied in which video frames are transmitted through extremely constrained network. The objective is to maximize the fluency and network efficiency. Furthermore, a crowd intelligence empowered dynamic adaptive video transmission method in extreme circumstances is presented. First, several typical ultra-low-bandwidth video transmission scenes are selected for network data collection. Second, working status data of modules internal devices are processed and evaluated. The evaluation method is empowered by crowd intelligence. Third, the reward function of Q-Learning is reconstructed to be dynamically updated according to device evaluation results and network status. Then, it can be corrected by exception handling strategies to adapt to the scenes better. Experiments on different scenes demonstrated the proposed method was effective, thereby verifying feasibility of the proposed method.