The real-time video analysis algorithm module based on tensorrt dynamic batch inference accesses the upper layer service at the model inference output end. Business involved:Video velocity measurement
[TOC]
model
The algorithm serves the underlying algorithm implementation : tracking algorithm, optical flow algorithm.
mutual
Interactive interface with the back-end : mainly related to grpc communication and shared memory read and write.
parallel
Parallel Task Management Module : Tasks that need to be processed in parallel can be placed here and currently include tracking tasks.
application
Business application module : including the underlying algorithm scheduling, business configuration, business table, etc.
Modify the underlying algorithm configuration, see “model/config/config.yaml”
De: model_path: './model/engine_file/model.432FP32.engine' input_name: 'inputs' output_name: 'outputs' cates: [ 'bg','water' ] half: False
pass in a json object. The configuration file format is detailed in "./application/config/water_flow_velocity_config.json"
[
{
"camera": "001",
"rois": 2,
"rois_shape": [[432, 131], [432, 131]],
"left_top": [[0, 0], [0, 0]],
"distance": 12.3,
"change": []
},
{
"camera": "002",
"rois": 1,
"rois_shape": [[432, 131]],
"left_top": [[0, 0]],
"distance": 11.0,
"change": [0]
}]
Enter the project root directory, business testing local test.
python module_test/manager_test.py
Start the grpc server and run python grpc/smserver.py