Method and system for maximizing uplink bandwidth in a communication system

1. A communication system comprising:

an end device generating an image stream;

a gateway coupled to the end device, the gateway including a gateway monitor agent collecting utilization rate data of the gateway, a task allocator, and an image inspector collecting inspection data from the received image stream;

an edge server coupled to the gateway via an uplink, the edge server including an edge server monitor agent collecting utilization rate data of the edge server, wherein the edge server and the gateway are each capable of performing an identical image inference task relating to processing the image stream; and

an analytics manager coupled to the gateway and the edge server, the analytics manager configured to determine an allocation strategy of the image inference task relating to processing the image stream to be performed by either the gateway or the edge server to maximize uplink bandwidth utilization from the gateway to the edge server, wherein the allocation strategy is selected based on a prediction of utilization rate determined by a utilization model trained by the collected utilization rate data from the gateway and the edge server, and wherein the task allocator employs the allocation strategy to assign the image inference task relating to processing the image stream to one of the gateway or the edge server.

2. The system of claim 1 , wherein the end device is one of a tablet, a mobile phone, a laptop computer, or a personal computer.

3. The system of claim 1 , wherein the image inspector compresses the image stream through identification of a key frame in the image stream.

4. The system of claim 3 , wherein the key frame is identified by a region of interest output by an image compression model trained by collected inspection data of previous image streams, and wherein images with key frames are extracted from the image stream and images without the region of interest are dropped.

5. The system of claim 4 , wherein the image inference task is determining the presence of a region of interest in an image of the image stream.

6. The system of claim 4 , wherein the image compression model is trained by a Convolution Neural Network (CNN) and an Auto-Encoder Network.

7. The system of claim 3 , wherein the edge server is operable to reconstruct the key frame in the image stream from low resolution to high resolution in real time to lower the uplink bandwidth utilization of the edge server.

8. The system of claim 1 , wherein the utilization rate data includes a bandwidth utilization rate and a real-time computation rate.

9. The system of claim 1 , wherein the utilization model includes a Recurrent Neural Network (RNN) and a Long Short Term Memory (LSTM) network that obtain a time-series prediction for bandwidth prediction.

10. The system of claim 1 , wherein the allocation strategy allocates the inference task to the edge server based on whether a computing workload value of the edge server and a timely bandwidth utilization value of the edge server are less than a threshold value.

11. A method of streamlining image processing in a communication system including a gateway coupled to an edge server having an edge server monitor agent, the gateway having a gateway monitor agent and an image inspector, wherein the edge server and the gateway are each capable of performing an identical image inference task relating to processing the image stream, the method comprising:

receiving an image stream from an end device;

collecting utilization rate data of the gateway via the gateway monitor agent;

collecting inspection data from the received image stream via the image inspector;

collecting utilization rate data of the edge server via the edge server monitor agent;

selecting an allocation strategy of the image inference task relating to processing the image stream to be performed by either the gateway or the edge server to maximize uplink bandwidth utilization from the gateway to the edge server, wherein the allocation strategy is selected based on a prediction of utilization rate determined by a utilization model trained by the collected utilization rate data from the gateway and the edge server via an analytics manager coupled to the gateway and the edge server; and

assigning the image inference task relating to processing the image stream to one of the gateway or the edge server based on the allocation strategy.

12. The method of claim 11 , wherein the end device is one of a tablet, a mobile phone, a laptop computer, or a personal computer.

13. The method of claim 11 , wherein the image inspector compresses the image stream through identification of a key frame in the image stream.

14. The method of claim 13 , further comprising training an image compression model through collected inspection data of previous image streams, wherein the key frame is identified by a region of interest output by the image compression model and wherein key frames are extracted from the image stream and images without the region of interest are dropped.

15. The method of claim 14 , wherein the image compression model is trained by a Convolution Neural Network (CNN) and an Auto-Encoder Network.

16. The method of claim 13 , further comprising reconstructing the key frame in the image stream from low resolution to high resolution in real time to lower the uplink bandwidth utilization of the edge server.

17. The method of claim 11 , wherein the utilization rate data includes a bandwidth utilization rate and a real-time computation rate.

18. The method of claim 11 , wherein the utilization model includes a Recurrent Neural Network (RNN) and a Long Short Term Memory (LSTM) network that obtain a time-series prediction for bandwidth prediction.