3 openVINO 运行 crossroad_camera_demo

参见官网https://docs.openvinotoolkit.org/latest/_inference_engine_samples_crossroad_camera_demo_README.html

模仿security_barrier_camera _demo.sh创建一个crossroad_camera_demo.sh，用于播放一段mp4视频，识别人员、人员属性、人员记数。我写的脚本在文章结尾处，模仿security_barrier_camera _demo.sh的目的是让大家知道openVINO大概的框架和运行、编译、步骤，各文件所在目录。

运行crossroad_camera_demo.sh脚本时

1.下载英特尔型号

下载模型到/ home / root1 / openvino_models / ir / FP32 /路径下

家用/目录root1 / openvino_models / IR / FP32 /安全/ object_detection /十字路口/ 0078 / dldt /

家用/目录root1 / openvino_models / IR / FP32 /安全/ OBJECT_ATTRIBUTES /行人/人的属性识别-十字路口-0230 / dldt /

家用/目录root1 / openvino_models / IR / FP32 /零售/ object_reidentification /行人/ rmnet_based / 0079 / dldt /

2.建立样本

编译/选择/英特尔/ openvino / inference_engine /样品/ crossroad_camera_demo

到/家庭/目录root1 / inference_engine_samples_build / Intel64位/发行/下

3.执行crossroad_camera_demo

./crossroad_camera_demo -d CPU -i /opt/intel/openvino/deployment_tools/demo/1.mp4 -m / home / root1 / openvino_models / ir / FP32 / Security / object_detection / crossroad / 0078 / dldt / person-vehicle-bike -detection-crossroad-0078.xml -m_pa /home/root1/openvino_models/ir/FP32/Security/object_attributes/pedestrian/person-attributes-recognition-crossroad-0230/dldt/person-attributes-recognition-crossroad-0230.xml -m_reid /home/root1/openvino_models/ir/FP32/Retail/object_reidentification/pedestrian/rmnet_based/0079/dldt/person-reidentification-retail-0079.xml

 

如果运行出现以下错误，一般都是环境变量地址错误引起的，先把环境变量脚本运行起来。

请参考如下处理步骤：

if [-e“$ ROOT_DIR /../../ bin / setupvars.sh”]; 然后

setupvars_path = “$ ROOT_DIR /../../斌/ setupvars.sh”

其他

printf“错误：找不到setupvars.sh \ n”

科幻

如果！。$ setupvars_path; 然后

printf "Unable to run ./setupvars.sh. Please check its presence. ${run_again}"

exit 1

fi

 

4、使用./crossroad_camera_demo -h 可查看参数

-h打印用法消息。

-i“”必需。视频或图像文件的路径。默认值为“cam”以与相机配合使用。

-m“”必需。人/车/自行车检测十字路口模型（.xml）文件的路径。

-m_pa“”可选。人员属性识别十字路口模型（.xml）文件的路径。

-m_reid“”可选。Person Reidentification零售模型（.xml）文件的路径。

-l“”可选。对于CPU自定义图层（如果有）。具有内核impl的共享库的绝对路径。

要么

-c“”可选。对于GPU定制内核，如果有的话。具有内核desc的xml文件的绝对路径。

-d“”可选。指定人员/车辆/自行车检测的目标设备（CPU，GPU，FPGA，HDDL，MYRIAD或HETERO）。

-d_pa“”可选。指定人员属性识别（CPU，GPU，FPGA，HDDL，MYRIAD或HETERO）的目标设备。

-d_reid“”可选。指定Person Reidentification Retail（CPU，GPU，FPGA，HDDL，MYRIAD或HETERO）的目标设备。

-pc可选。启用每层性能统计信息。

-r可选。输出推断结果为原始值。

-t可选。人/车/自行车十字路口检测的概率阈值。

-t_reid可选。用于人重新识别的两个向量之间的余弦相似性阈值。

-no_show可选。没有显示处理过的视频

-auto_resize可选。支持可调整大小的输入，支持ROI裁剪和自动调整大小。

 

./crossroad_camera_demo -d CPU

-i /opt/intel/openvino/deployment_tools/demo/1.mp4

-m /home/root1/openvino_models/ir/FP32/Security/object_detection/crossroad/0078/dldt/person-vehicle-bike-detection-crossroad-0078.xml

-m_pa /home/root1/openvino_models/ir/FP32/Security/object_attributes/pedestrian/person-attributes-recognition-crossroad-0230/dldt/person-attributes-recognition-crossroad-230.xml

-m_reid /home/root1/openvino_models/ir/FP32/Retail/object_reidentification/pedestrian/rmnet_based/0079/dldt/person-reidentification-retail-0079.xml

 

5 以下贴出脚本完整代码，创建一个crossroad_camera_demo.sh，把以下代码粘贴进去。

#!/usr/bin/env bash

 

target="CPU"

target_precision="FP32"

 

ROOT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"

target_image_path="$ROOT_DIR/1.mp4"

 

if [ -e "$ROOT_DIR/../../bin/setupvars.sh" ]; then

setupvars_path="$ROOT_DIR/../../bin/setupvars.sh"

else

printf "Error: setupvars.sh is not found\n"

fi

 

if ! . $setupvars_path ; then

printf "Unable to run ./setupvars.sh. Please check its presence. ${run_again}"

exit 1

fi

 

# Step 1. Downloading Intel models

printf "Downloading Intel models\n"

 

downloader_path="${INTEL_OPENVINO_DIR}/deployment_tools/tools/model_downloader/downloader.py"

models_path="$HOME/openvino_models/ir/${target_precision}"

 

person_attributes_recognition_crossroad=person-attributes-recognition-crossroad-0230

person_vehicle_bike_detection_crossroad=person-vehicle-bike-detection-crossroad-0078

person_reidentification_retail=person-reidentification-retail-0079

 

person_attributes_recognition_crossroad_path=${models_path}/Security/object_attributes/pedestrian/${person_attributes_recognition_crossroad}/dldt/${person_attributes_recognition_crossroad}

person_vehicle_bike_detection_crossroad_path=${models_path}/Security/object_detection/crossroad/0078/dldt/${person_vehicle_bike_detection_crossroad}

person_reidentification_retail_path=${models_path}/Retail/object_reidentification/pedestrian/rmnet_based/0079/dldt/${person_reidentification_retail}

 

# 下载模型

if ! [ -f "${person_attributes_recognition_crossroad_path}.xml" ] && ! [ -f "${person_attributes_recognition_crossroad_path}.bin" ]; then

printf "\nRun $downloader_path --name $person_attributes_recognition_crossroad --output_dir ${models_path}\n\n"

$python_binary $downloader_path --name $person_attributes_recognition_crossroad --output_dir ${models_path}

else

printf "\n${person_attributes_recognition_crossroad} have been loaded previously, skip loading model step."

fi

 

if ! [ -f "${person_vehicle_bike_detection_crossroad_path}.xml" ] && ! [ -f "${person_vehicle_bike_detection_crossroad_path}.bin" ]; then

printf "\nRun $downloader_path --name $person_vehicle_bike_detection_crossroad --output_dir ${models_path}\n\n"

$python_binary $downloader_path --name $person_vehicle_bike_detection_crossroad --output_dir ${models_path}

else

printf "\n${person_vehicle_bike_detection_crossroad} have been loaded previously, skip loading model step."

fi

 

if ! [ -f "${person_reidentification_retail_path}.xml" ] && ! [ -f "${person_reidentification_retail_path}.bin" ]; then

printf "\nRun $downloader_path --name $person_reidentification_retail --output_dir ${models_path}\n\n"

$python_binary $downloader_path --name $person_reidentification_retail --output_dir ${models_path}

else

printf "\n${person_reidentification_retail} have been loaded previously, skip loading model step.\n\n"

fi

 

# Step 2. Build samples

printf "Build Inference Engine samples\n"

 

samples_path="${INTEL_OPENVINO_DIR}/deployment_tools/inference_engine/samples"

 

#是否存在cmake命令

if ! command -v cmake &>/dev/null; then

printf "\n\nCMAKE is not installed. It is required to build Inference Engine samples. Please install it. ${run_again}"

exit 1

fi

 

# demo所在目录

OS_PATH=$(uname -m)

NUM_THREADS="-j2"

 

if [ $OS_PATH == "x86_64" ]; then

OS_PATH="intel64"

NUM_THREADS="-j8"

fi

 

build_dir="$HOME/inference_engine_samples_build"

if [ -e $build_dir/CMakeCache.txt ]; then

    rm -rf $build_dir/CMakeCache.txt

fi

mkdir -p $build_dir

cd $build_dir

 

# 编译demo

# cmake -DCMAKE_BUILD_TYPE=Release $samples_path

# make $NUM_THREADS crossroad_camera_demo

 

# Step 3. Run samples

binaries_dir="${build_dir}/${OS_PATH}/Release"

cd $binaries_dir

 

# ./crossroad_camera_demo -d CPU

# -i /opt/intel/openvino/deployment_tools/demo/1.mp4

# -m /home/root1/openvino_models/ir/FP32/Security/object_detection/crossroad/0078/dldt/person-vehicle-bike-detection-crossroad-0078.xml

# -m_pa /home/root1/openvino_models/ir/FP32/Security/object_attributes/pedestrian/person-attributes-recognition-crossroad-0230/dldt/person-attributes-recognition-crossroad-0230.xml

# -m_reid /home/root1/openvino_models/ir/FP32/Retail/object_reidentification/pedestrian/rmnet_based/0079/dldt/person-reidentification-retail-0079.xml

 

printf "Run ./crossroad_camera_demo -d $target -i $target_image_path -m "${person_vehicle_bike_detection_crossroad_path}.xml" -m_pa "${person_attributes_recognition_crossroad_path}.xml" -m_reid "${person_reidentification_retail_path}.xml" \n\n"

./crossroad_camera_demo -d $target -i $target_image_path -m "${person_vehicle_bike_detection_crossroad_path}.xml" -m_pa "${person_attributes_recognition_crossroad_path}.xml" -m_reid "${person_reidentification_retail_path}.xml"

 

printf "Demo completed successfully.\n\n"

 

# -i“”必需。视频或图像文件的路径。默认值为“cam”以与相机配合使用。

# -m“”必需。人/车/自行车检测十字路口模型（.xml）文件的路径。

# -m_pa“”可选。人员属性识别十字路口模型（.xml）文件的路径。

# -m_reid“”可选。Person Reidentification零售模型（.xml）文件的路径。

# -l“”可选。对于CPU自定义图层（如果有）。具有内核impl的共享库的绝对路径。

# 要么

# -c“”可选。对于GPU定制内核，如果有的话。具有内核desc的xml文件的绝对路径。

# -d“”可选。指定人员/车辆/自行车检测的目标设备（CPU，GPU，FPGA，HDDL，MYRIAD或HETERO）。

# -d_pa“”可选。指定人员属性识别（CPU，GPU，FPGA，HDDL，MYRIAD或HETERO）的目标设备。

＃-d_reid“”可选。指定人员重新识别零售（CPU，GPU，FPGA，HDDL，MYRIAD或HETERO）的目标设备。

＃-pc可选。启用每层性能统计信息。

＃-r可选。输出推断结果为原始值。

＃-t可选。人/车/自行车十字路口检测的概率阈值。

＃-t_reid可选。用于人重新识别的两个向量之间的余弦相似性阈值。

＃-no_show可选。没有显示处理过的视频

＃-auto_resize可选。支持可调整大小的输入，支持ROI裁剪和自动调整大小。