Skip to content

AIS-Bonn/JetsonTRTPerception

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

JetsonTRTPerception

In this repo we publish the inference code running on the Jetson NX Smart Edge Sensors for the paper:
3D Semantic Scene Perception using Distributed Smart Edge Sensors

Citation

Simon Bultmann and Sven Behnke:
3D Semantic Scene Perception using Distributed Smart Edge Sensors
Accepted for 17th International Conference on Intelligent Autonomous Systems (IAS), Zagreb, Croatia, June 2022.

Installation

Dependencies

The code was tested with ROS melodic and Ubuntu 18.04.

On the host PC, TensorRT is required, install it via the official documentation. The code was tested with:
Host PC: TensorRT 7.2.3-1+cuda11.1
Jetson NX: Jetpack 4.5.1 (TensorRT 7.1.3, CUDA 10.2)

The code is not currently compatible with TensorRT 8.x, as the UFF-Parser is used for RGB detection and pose estimation models.

We depend on a custom version of TensorRT Open Source Software (OSS), included in this repo. The gridAnchor and flattenConcat plugins have been updated to support dyncamic shapes, and the gridAnchor plugin has been fixed to support rectangular input images.

ROS packages

Clone this repo and the SmartEdgeSensor3DScenePerception repo inside your catkin workspace:

cd catkin_ws/src
git clone https://github.com/AIS-Bonn/JetsonTRTPerception.git
git clone https://github.com/AIS-Bonn/SmartEdgeSensor3DScenePerception.git

Build TensorRT OSS:

cd JetsonTRTPerception/TensorRT
mkdir -p build && cd build

export TRT_LIBPATH=/usr/lib/x86_64-linux-gnu/
cmake .. -DTRT_LIB_DIR=$TRT_LIBPATH -DTRT_OUT_DIR=`pwd`/out 

make -j$(nproc)

For furhter build instructions (Native build on Jetson / cross-compile for Jetson) see: https://github.com/NVIDIA/TensorRT#building-tensorrt-oss

Build the ros packages:

cd ..
catkin build --cmake-args -DCMAKE_BUILD_TYPE=Release
source devel/setup.bash

Launch

roslaunch jetson_trt_pose pose_estimation.launch feedback:=skel camera:=d455

color image topic: /d455/color/image_raw
color camera info: /d455/color/camera_info
depth image topic: /d455/depth/image_rect_raw
depth camera info: /d455/depth/camera_info

feedback topic: /d455/skel_pred

If thermal detector model is given:
thermal image topic: /d455/lepton/image
thermal camera info: /d455/lepton/camera_info