Trial record of ModelBox end cloud collaborative AI development kit (rk3568) (III)

This part is ： Use yolov3 Do mask test

---

Mask detection is different from vehicle detection , Vehicle detection is built into SDK in , Model file 、 Relevant configurations are provided as solutions , The created project is there . The mask detection needs to download the model and configuration files by yourself , And put it into the corresponding project directory , In fact, it is also very convenient .
Download address , Extraction code modbox choice common.zip（ Functional units and configurations 、 Task allocation 、 Test video 5M） and rknpu2.zip（ Reasoning function unit and process configuration 14M） Download it .

Still use it first create.py Create a mask_det_yolo3 The engineering of , And then use -t infer Option to create a named mask_infer Reasoning function unit of , It'll be in the engineering catalog model The following content is generated under

|-- model|   `-- mask_infer|       `-- mask_infer.toml

take rknpu2.zip Inside rknn Model files and 2 Configuration files are put into （ Covering the original mask_infer.toml）：

workspace/mask_det_yolo3/model/mask_infer/|-- mask_infer.toml|-- yolo3_resnet18_mask_det_288x512.toml`-- yolo3_resnet18_mask_det_288x512_rknpu2.rknn

mask_infer.toml The output of three models is defined in , It should be for head Head shoulder 、face、mask masks Information about .

Next, use parameters -t python establish 2 A functional unit , Namely yolo3_post post-processing , and draw_mask_bbox Draw a frame on the original picture , The generated content is as follows ：

workspace/mask_det_yolo3/etc/`-- flowunit    |-- cpp    |-- draw_mask_bbox    |   |-- draw_mask_bbox.py    |   `-- draw_mask_bbox.toml    |-- readme.txt    `-- yolo3_post        |-- yolo3_post.py        `-- yolo3_post.toml

take common.zip Copy the contents of the functional units inside to overwrite the original file .

Display of results with or without masks , Its logic is realized in the drawing unit ： If face Detection box and mask Detection box Coincidence Greater than a certain threshold , It was sentenced to wear a mask ; otherwise , It will be judged as not wearing a mask ; If you don't detect face Check box , Will show Unknown.

The next step is Serial flow chart （graph/mask_det_yolo3.toml）, and Configure the input and output of the task （bin/mock_task.toml）, Of course, there are already configured files in the download , Copy to the corresponding directory . At the same time, copy the test video to data Catalog .

mask_det_yolo3.toml It's defined in ：
The processing flow of each functional unit

data_source_parser > video_demuxer > video_decoder > image_resize > mask_detection > yolo3_post > draw_mask_bbox > video_out

as well as Interface docking configuration between output and input between functional units , Don't talk about the details .

The input and output of the configuration task are hello_world And the vehicle inspection department , I won't say .

This is the end of preparations , The following is to enter the project directory , Conduct build_project.sh then bin/main.sh Run the application to detect the wearing of masks in video files , And output the results to the video file .

There is another one in the download camera Configuration file for , Input and output can be configured as video streams , Try again later .