Shengteng Atlas 200I DK A2 realizes helmet recognition

Article directory

environment dependence
compile
test
Summarize

environment dependence

software	Version	illustrate	method of obtaining
mxVision	5.0.RC2	mxVision package	method of obtaining
Ascend-CANN-toolkit	6.2.RC2	Ascend-cann-toolkit development kit package	method of obtaining
Ubuntu	22.04

Code warehouse address:
https://gitee.com/ascend/ascend_community_projects/tree/310B/HelmetIdentification_V2

Mirror version:
Insert image description here

compile

获取模型：https://gitee.com/link?target=https%3A%2F%2Fmindx.sdk.obs.cn-north-4.myhuaweicloud.com%2Fmindxsdk-referenceapps%2520%2Fcontrib%2FHelmetIdentification%2Fmodel.zip

Unzip model.zip and move the onnx file to the model directory of the project:
Insert image description here

Conversion model: Note that the atc-env.sh conversion script in the source code is not used here.
Configure environment variables:

source /usr/local/Ascend/ascend-toolkit/set_env.sh
source /usr/local/Ascend/mxVision-5.0.RC2/set_env.sh

Transfer model:

cd ~/HelmetIdentification_V2/model
atc --model=./YOLOv5_s.onnx --framework=5 --output=YOLOv5_s --insert_op_conf=./aipp_YOLOv5.config --input_format=NCHW --log=info --soc_version=Ascend310B1 --input_shape="images:1,3,640,640"

After executing the above command line, the model model will appear in the directory om
Insert image description here

Modify CMakeList.txt

cd ~/HelmetIdentification_V2/src

Change the 24 and 35 lines to in the /usr/local/Ascend/ascend-toolkit/latest/aarch64-linux directory and directory, you need to change it to your own directoryincludelib64toolkit
Insert image description here

Compilation preparation

Update installation library files

apt-get update 
apt-get install -y libavformat-dev

Create soft links to three libraries

ln -s /usr/lib/aarch64-linux-gnu/libavcodec.so.58 /usr/lib/aarch64-linux-gnu/libavcodec.so
ln -s /usr/lib/aarch64-linux-gnu/libavutil.so.56 /usr/lib/aarch64-linux-gnu/libavutil.so
ln -s /usr/lib/aarch64-linux-gnu/libavformat.so.58 /usr/lib/aarch64-linux-gnu/libavformat.so

Since the code provides the difference between video and , here we choose . After backing up the two files, leaveimagemain.cppvideomain.cpp

cd ~/HelmetIdentification_V2/src
rm main-image.cpp
cd ..
mkdir build_video
cd build_video
cmake ..

Insert image description here

make -j4

Insert image description here
At this point, the executable file will be generated in the ~/HelmetIdentification_V2 directory.main

test

Under the HelmetIdentification_V2 folder, you need to create a new result folder and two internal one、two folders for Store results

cd ~/HelmetIdentification_V2
mkdir result
cd result
mkdir one
mkdir two

Return toHelmetIdentification_V2 folder and run the following command

./main test_person.h264 1920 1080

1920 1080 is the width and height of the input video
Insert image description here
The resulting image is saved in the folderHelmetIdentification_V2/result, and the folderone saves the first The results of the first input, the foldertwo saves the results of the second input.

The output sample is:
The red box marked should be the information that the helmet is not worn.

Summarize

This article can be regarded as an introductory example of the Shengteng series. Once you run through it, you should be able to know the general operation process. What follows should be another unforgettable development journey. I hope you can hold on! ! !

Reference documents:
https://gitee.com/ascend/ascend_community_projects/tree/310B/HelmetIdentification_V2
https://zhuanlan.zhihu .com/p/652517700
If reading this article is useful to you, please like and save it! ! !
November 24, 2023 14:55:12
Insert image description here