Finding freedom: Freespace segmentation using occupancy grids from stereo disparities

This project presents an implementation of the freespace segmentation approach proposed in [1]. This segmentation approach uses an occupancy grid representation which only requires a stereo disparity image and associated camera model for freespace computation, making it more computationally efficient than some state-of-the-art approaches using CNNs. Different segmentation algorithms were evaluated and a Kalman filter was implemented for improving the quality of disparity images. The performance of the implementation was evaluated on a labelled portion of the KITTI dataset based on precision and recall and demonstrates promising results, outperforming the state-of-the-art approaches in some metrics including runtime performance. However, the performance of the system is dependant on the quality of the disparity images provided and performs poorly in certain environments.