Feat: People detection with Region of Interest

.gitignore

@@ -0,0 +1 @@
+models/

download_models/downloadModels.sh

@@ -39,6 +39,7 @@ if [ "$REFRESH_MODE" -eq 1 ]; then
 fi
 
 pipelineZooModel="https://github.com/dlstreamer/pipeline-zoo-models/raw/main/storage/"
+openVINOModelZoo="https://storage.openvinotoolkit.org/repositories/open_model_zoo/2022.3/models_bin/1/"
 
 # $1 model file name
 # $2 download URL
@@ -136,8 +137,22 @@ downloadTextRecognition() {
     fi
 }
 
+downloadPersonDetection() {
+    personDetection0106="person-detection-0200"
+    modelType="object_detection"
+    personjsonModelPath="$modelType/$personDetection0106/$modelPrecisionFP16INT8/$personDetection0106.bin"
+    echo $personjsonModelPath
+
+    if [ ! -f $personjsonModelPath ]; then
+        getModelFiles $personDetection0106 $openVINOModelZoo$personDetection0106 $modelPrecisionFP16INT8 $modelType                
+    else
+        echo "personDetection0106 $modelPrecisionFP16INT8 model already exists, skip downloading..."
+    fi
+}
+
 ### Run custom downloader section below:
 downloadYolov5sFP16INT8
 downloadEfficientnetb0
 downloadHorizontalText
 downloadTextRecognition
+downloadPersonDetection

src/Dockerfile

@@ -4,7 +4,7 @@
 # SPDX-License-Identifier: Apache-2.0
 #
 
-FROM intel/dlstreamer:2024.1.1-dev-ubuntu22 as base
+FROM intel/dlstreamer:2024.1.2-ubuntu24 as base
 USER root
 
 FROM base as build-default

src/docker-compose.yml

@@ -47,6 +47,9 @@ services:
       - INPUTSRC=${INPUTSRC:-rtsp://localhost:8554/camera_0}
       - RENDER_MODE=${RENDER_MODE:-0} #RENDER_MODE=1 will work only after running xhost +local:docker
       - DISPLAY=$DISPLAY
+      - HTTP_PROXY=${HTTP_PROXY}
+      - HTTPS_PROXY=${HTTPS_PROXY}
+      - NO_PROXY=${NO_PROXY}
     volumes:
       - ${RESULTS_DIR:-../results}:/tmp/results
       - ~/.Xauthority:/home/dlstreamer/.Xauthority

src/entrypoint.sh

@@ -32,7 +32,7 @@ DECODE="${DECODE:="decodebin force-sw-decoders=1"}" #decodebin|vaapidecodebin
 DEVICE="${DEVICE:="CPU"}" #GPU|CPU|MULTI:GPU,CPU
 
 show_help() {
-	echo "usage: \"--pipeline_script_choice\" requires an argument yolov5s.sh|yolov5s_effnetb0.sh|yolov5s_full.sh"
+	echo "usage: \"--pipeline_script_choice\" requires an argument yolov5s.sh|yolov5s_effnetb0.sh|yolov5s_full.sh|people_detection.sh"
 }
 
 while :; do
@@ -66,7 +66,7 @@ while :; do
 
 done
 
-if [ "$PIPELINE_SCRIPT" != "yolov5s.sh" ] && [ "$PIPELINE_SCRIPT" != "yolov5s_effnetb0.sh" ] && [ "$PIPELINE_SCRIPT" != "yolov5s_full.sh" ]
+if [ "$PIPELINE_SCRIPT" != "yolov5s.sh" ] && [ "$PIPELINE_SCRIPT" != "yolov5s_effnetb0.sh" ] && [ "$PIPELINE_SCRIPT" != "yolov5s_full.sh" ] && [ "$PIPELINE_SCRIPT" != "people_detection.sh" ]
 then
 	echo "Error on your input: $PIPELINE_SCRIPT"
 	show_help
@@ -96,11 +96,15 @@ elif grep -q "file" <<< "$INPUTSRC"; then
 	arrfilesrc=(${INPUTSRC//:/ })
 	# use vids since container maps a volume to this location based on sample-media folder
 	inputsrc="filesrc location=vids/"${arrfilesrc[1]}" ! qtdemux ! h264parse "
-elif grep -q "video" <<< "$INPUTSRC"; then
+elif grep -q "/dev" <<< "$INPUTSRC"; then
 	inputsrc="v4l2src device="$INPUTSRC
 	DECODE="$DECODE ! videoconvert ! video/x-raw,format=BGR"
-	# when using realsense camera, the dgpu.0 not working
+	# for http:// videos	
+elif grep -q "http" <<< "$INPUTSRC"; then
+	inputsrc="urisourcebin buffer-size=4096 uri="$INPUTSRC
+	DECODE="decodebin"
 else
+	# when using realsense camera, the dgpu.0 not working
 	# rs-serial realsenssrc
 	inputsrc="realsensesrc cam-serial-number="$INPUTSRC" stream-type=0 align=0 imu_on=false"
 	echo "----- in run_gst.sh COLOR_WIDTH=$COLOR_WIDTH, COLOR_HEIGHT=$COLOR_HEIGHT, COLOR_FRAMERATE=$COLOR_FRAMERATE"

src/pipelines/people_detection.sh

@@ -0,0 +1,27 @@
+ #!/bin/bash
+#
+# Copyright (C) 2024 Intel Corporation.
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+
+PRE_PROCESS="${PRE_PROCESS:=""}" #""|pre-process-backend=vaapi-surface-sharing|pre-process-backend=vaapi-surface-sharing pre-process-config=VAAPI_FAST_SCALE_LOAD_FACTOR=1 
+AGGREGATE="${AGGREGATE:="gvametaaggregate name=aggregate !"}" # Aggregate function at the end of the pipeline ex. "" | gvametaaggregate name=aggregate
+PUBLISH="${PUBLISH:="name=destination file-format=2 file-path=/tmp/results/r$cid\"_gst\".jsonl"}" # address=localhost:1883 topic=inferenceEvent method=mqtt
+
+ROI="${ROI:="0,0,400,400"}"
+
+if [ "$RENDER_MODE" == "1" ]; then
+    OUTPUT="${OUTPUT:="! videoconvert ! video/x-raw,format=I420 ! gvawatermark ! videoconvert ! fpsdisplaysink video-sink=ximagesink sync=true --verbose"}"
+else
+    OUTPUT="${OUTPUT:="! fpsdisplaysink video-sink=fakesink sync=true --verbose"}"
+fi
+
+echo "decode type $DECODE"
+echo "Run yolov5s pipeline on $DEVICE with batch size = $BATCH_SIZE"
+
+gstLaunchCmd="GST_DEBUG=\"GST_TRACER:7\" GST_TRACERS=\"latency_tracer(flags=pipeline,interval=100)\" gst-launch-1.0 $inputsrc ! $DECODE ! gvaattachroi roi=$ROI ! gvadetect batch-size=$BATCH_SIZE model-instance-id=odmodel name=detection model=models/object_detection/person-detection-0200/FP16-INT8/person-detection-0200.xml threshold=.5 device=$DEVICE $PRE_PROCESS inference-region=1 ! $AGGREGATE gvametaconvert name=metaconvert add-empty-results=true ! gvametapublish name=destination file-format=2 file-path=/tmp/results/r$cid\"_gst\".jsonl $OUTPUT 2>&1 | tee >/tmp/results/gst-launch_$cid\"_gst\".log >(stdbuf -oL sed -n -e 's/^.*current: //p' | stdbuf -oL cut -d , -f 1 > /tmp/results/pipeline$cid\"_gst\".log)"
+
+echo "$gstLaunchCmd"
+
+eval $gstLaunchCmd

src/res/person-detection.env

@@ -0,0 +1 @@
+ROI="400,200,600,400"