A boilerplate of consumer-producer where the producer is a camera capture.
There are two versions: a queue and a shared memory version. The difference is the way the camera process sends the frame. The queue version has less stuff to manage, while the shared version has some conditions and lock to handle due to resource race conditions. Queue version has a higher frame rate (tested with 3 cameras) but some cameras may send more frames than others. The shared memory version has a lower frame rate, but each camera sends approximately the same number of frames.
The engine_manager.py makes the interface with the external world. In this file you should implement functions to be accessed externally via gRPC, HTTP request, function calls, or whatever you want.
Also, in this file you'll need to specify a list of consumers you want to have, just open the file and follow the example.
A function to add a new camera is already implemented, and you can follow the example to add more features.
The consumer_manager.py will be instantiated by the EngineManager, and its job is to handle the data received from the producers, sending to all consumers. Every consumer will receive the same data item, and after processing it, each result of each consumer will be placed together in a queue, which can be accessed via EngineManager.
The consumers must be implemented following the examples, and must inherit multiprocessing.Process and implement the run() function.
In the queue version:
Lastly, camera.py capture the camera frames and put in a queue if is not full.
While in the shared memory version:
Lastly, camera.py capture the camera frames and put in a shared memory.
Every class in this code when instantiated will spawn a new process. Here the multiprocessing library is used to achieve real parallelism, escaping the python GIL, and OpenCV to capture the camera frames.
Install opencv-python and clone this repo.
Edit the URL at engine_manager.py file and python engine_manager.py.
This code is intended to be efficient. If you have tips or any comment that can benefit this code, please, do not hesitate to open an issue or a pull request.