Background of thesis project
Volvo Construction Equipment (VCE) is one of the world's largest manufacturers of construction equipment. With enhancing the artificial intelligent area and learning-based approaches, there is a big trend for shifting towards autonomous vehicles which can localize themselves in the environment, perceive and plan. In the construction domain, 3D object detection is an essential mean to enable autonomous driving where machines can detect objects and track them. Deep neural networks (DNNs) provide state-of-the-art results for 3D object detection and segmentation [1, 2]. However, while DNNs can rely on public general-purpose training datasets , such datasets are still mostly lacking for construction environments. On the other hand, preparing the dataset is very time consuming and requires huge effort of labelling. To maintain a leading position, it is essential for Volvo to develop innovative and cost-efficient and high-performance solutions for reducing the labling effort of point cloud datasets.
This master thesis is suitable for one student that is completing their studies in computer science. The thesis will be lead by Volvo Construction Equipment. Desired start date is Jan 2023.
Description of thesis work
Prior studies [4, 5] tried to solve the problem of automatic point cloud labling; however, they are inaccurate for off-road environments.
This project aims to implement an efficient automated framework for labling point cloud data for object detection and semantic segmentation tasks. The original dataset will be provided by VCE containing point cloud video frames obtained by the LiDAR sensor. The student is preferably knowledgeable in deep learning, and Python is meriting. The main activities are:
- Reviewing related studies and commercial tools for point cloud auto-labling tool.
- Implementing an efficient point cloud auto-labling framework for off-road environments.
- Compare the accuracy and the cost of using the proposed framework with the manual labling manner.
A full report of the work carried out will be prepared and presented to staff at Volvo CE. It will be required for the student to perform the thesis at Volvo CE facilities in Eskilstuna.
 Zhou, Yin, and Oncel Tuzel. "Voxelnet: End-to-end learning for point cloud based 3d object detection." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.
 Aygun, Mehmet, et al. "4d panoptic lidar segmentation." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021.
 Behley, Jens, et al. "Towards 3D LiDAR-based semantic scene understanding of 3D point cloud sequences: The SemanticKITTI Dataset." The International Journal of Robotics Research 40.8-9 (2021): 959-967.
 Bloembergen, Daan, and Chris Eijgenstein. "Automatic labelling of urban point clouds using data fusion." arXiv preprint arXiv:2108.13757 (2021).
 Li, MingHui, and Yanning Zhang. "3D Point Cloud Labeling Tool for Driving Automatically." 2020 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). IEEE, 2020.
Thesis Level: Master
Starting date: January 2023
Number of students: one student
Name, title, phone
Sara Afshar, Research Owner at Emerging Technology, +46 1654 15707
Mohammad Loni, Research Engineer AI at Emerging Technology, +46 165414570