Obstacles Detector for Self-Driving Cars

This repo contains a ROS node to detect cars and pedestrian in a self-driving car with LiDAR/radar/camera output.

lidar_node

The node takes the PointCloud from the LIDAR, and does a series of operations to help the detection:

• ground plane segmentation and removal
• removing points above and below certain heights (z-value), and those corresponding to the capture vehicle
• performing clustering via a Breadth-First Search, and computing the centroid of each cluster found

All of the real code currently resides in /src/lidar/src/lidar_node.cpp.

To use this node, first build using catkin-make.

camera

The detection of the camera frames is done using Deep Learning, specifically the Faster-RCNN library.

Fusion

Using the output from the camera, we can use the radar and LIDAR to get a more precise detection. Also, a Kalman filter is used to get more accurate results.