A mapping and obstacle avoidance project using 2D LiDAR scanner and Kinect RGB-D camera
ROS nodes : LiDAR scanning algorithm node and for the Kinect perception node.
LiDAR Node (in folder my_first) subscribes to the laser_scan message published by rplidar_ros package (You will need to install it in your catkin workspace). The Node will apply a median filter to laser scan, followed by generation of a point cloud of the indoor environment (a corridor, for example). Using this point cloud, an angle with respect to current driving direction is calculated and corresponding wheel velocities are generated using a PID controller. The Kinametic equations may change with respect to the dimensions of the robot.
The Kinect Node, uses Openni package do publish the raw_image message in ROS. This is subscribed by the package built for Kinect perception. A caffe model for Google's MobileNet CNN is used for detecting objects. This is followed by a Single Shot Detector (SSD) Network which determines the locations of the objects in the image. Corresponding depth images from Kinect camera are used to determine the closest objects and driving angle is caluclated using a PID controller.
This implementation runs in real-time.
The prerequisites are added in the prereqs compressed folder. It contains the rplidar_ros package, openni_camera and other useful packages which will be used in running the main nodes.
Please note that the sensors work individually. It is an experimentation for both sensors. No sensor fusion is used in this project as yet.