Occupancy Grid Map From Monocular Camera Frame

This repository hosts MATLAB script for generating occupancy grid map from a monocular camera frame. The trained DNN is attached separately with this release.

The monocular camera frame (first person view) is passed to a deep neural network trained to estimate free space around a vehicle using semantic segmentation. Free space estimation identifies areas in the environment where the ego vehicle can drive without hitting any obstacles such as pedestrians, curbs, or other vehicles. A vehicle can use a variety of sensors to estimate free space such as RADAR, LiDAR, or cameras. This project focuses on estimating free space from a monocular camera using semantic segmentation and creating an occupancy grid map using the same. This occupancy grid map can be used to create a vehicle costmap, which can be then used for path planning and navigation.

Results

Result	Image
Test Frame
First Person View (FPV)
Free Space Prediction [FPV]
Free Space Confidence Scores [FPV]
Bird's Eye View (BEV)
Free Space Confidence Scores [BEV]
Occupancy Grid Map