Stereo vision based obstacle mapping

** Don't star, this is not my active repo!** Currently it is under a private repository.

Original plan is to merge the two inputs of obstacle bounding boxes and generate a optimized obstacle cubicle map.

First try to reproduce the result of paper: Stereo Vision-based Semantic 3D Object and Ego-motion Tracking for Autonomous Driving of ECCV 2018.

@article{li2018stereo,
  title={Stereo Vision-based Semantic 3D Object and Ego-motion Tracking for Autonomous Driving},
  author={Li, Peiliang and Qin, Tong and Shen, Shaojie},
  journal={arXiv preprint arXiv:1807.02062},
  year={2018}
}

The semantic_mapping_node subscribes the obstacle map information obstacle_detection::MapInfo from two sets of cameras and pose information geometry_msgs::PoseStamped /ugv_slam_node/posestamped from wide camera.

Prerequisite

The package is based on Ubuntu 16.04 with ROS Kinetics. Additionally you should install:

1. OPENCV 3.1+

2. Eigen 3.2+

3. Ceres Solver

4. Install VSLAM and object detection repositories in <YOUR_CATKIN_WORKSPACE>:

Installation

    cd catkin_ws/src

    git clone git@github.com:zhanghanduo/stereo_semantic_mapping.git

    cd ..

    catkin_make

Demo

    roslaunch semantic_mapping demo.launch

RoadMap

• [] Early Development - Mapping without optimization
• [] Object Tracking
• [] Sparse Feature Observation
• [] Semantic 3D Object Measurement
• [] Point Cloud Alignment
• [] Joint Optimization
• [] Evaluation