Reusability of 3D Feature Based Maps for Automobiles (Bachelor's Final Year Project)
Reusability of 3D Feature Based Maps for Automobiles (Bachelor's Final Year Project)
Embedded Files
The proposed model uses a stereo camera as sensor with ORB feature detection and SLAM algorithm to create a universal 3D map of the environment that can be used for most of the environmental conditions, unlike normal 3D maps that can only be used for that specific environmental condition, thereby increasing the accuracy of the localization. The system was embedded with a self-made conventional algorithm which aimed to accurately identify the number of cars present before and not present now with their positions and vice versa. The whole system, with some amendments, can be used by AV industries to help them develop their autonomous vehicles.
The proposed model uses a stereo camera as sensor with ORB feature detection and SLAM algorithm to create a universal 3D map of the environment that can be used for most of the environmental conditions, unlike normal 3D maps that can only be used for that specific environmental condition, thereby increasing the accuracy of the localization. The system was embedded with a self-made conventional algorithm which aimed to accurately identify the number of cars present before and not present now with their positions and vice versa. The whole system, with some amendments, can be used by AV industries to help them develop their autonomous vehicles.
The first image is the image of ZED stereo camera that we used to collect the data. The data was then processed through MOBILENETV2 neural network for image segmentation and finally the data was fed to ORB and SLAM algorithm for real world map creation (mapping) and localization. The output is show below
The left blue points are raw car points extracted by SLAM algorithm and right multicolor points are refined, outlier removed car points (done using self-made conventional algorithm and DBSCAN). The red boxes denote detected cars and blue boxes denotes the ground truth of those cars. The approximate difference between the estimated positions of the cars to actual positions was about 50 cm from center to center.
A Detailed Video Description
A Detailed Video Description
reusability of 3D feature based maps for automobiles_demo.mp4
Page updated
Google Sites
Report abuse