Thesis supervisor: Levente Hajder
Location of studies (in Hungarian): ELTE IK Abbreviation of location of studies: ELTE
Description of the research topic:
The development of the functions of a self-driving car is currently taking place in several areas (sensor perception, fusion behavior planning, intervention) in parallel. To achieve the highest level of self-driving, signals from various sensory devices, such as video, radar, LiDAR, acceleration sensors, gyroscopes, GPS signals help the vehicle to map the environment and identify obstacles. Based on these inputs, it is possible to plan and intervene in the vehicle's progress.
To increase the accuracy of the representation of the environment, it is worth combining the signals of the different sensors. The fused signal can be used to highlight the advantageous properties of the sensors in some cases, while the disadvantages can be pushed into the background for some cases. There are several solutions in the literature for environmental representation achieved by the fusion of video and LiDAR / radar signals, but the use case solutions are still incomplete for video and radar and video and LiDAR fusion cases. The production of such a data set is necessary for the development of the autonomous driving function.
The aim of this dissertation is to create an algorithm that generates real LiDAR / radar echoes and spectra synthetically using video. The resulting data sets can then be used to simulate and validate the self-driving function. Deep Learning architectures (such as Generative Adversarial Network GAN) could be used to generate such synthetic data. The aim of the dissertation is also to validate the result, i.e. whether the resulting point clouds successfully and accurately represent the environment. Publicly available datasets (NuScenes) exist for the task, but industrial data sources can also be used in the dissertation. Although the aim of the research is to create models and algorithms, the results of the dissertation can be used directly in the ongoing industrial projects and developments, with which they can be used directly to implement the self-driving functionality at the highest level.
Required language skills: English Recommended language skills (in Hungarian): Hungarian Further requirements: SW: python (Numpy, Pandas, Matplotlib), machine learning: TensorFlow, SKLearn, Pytorch, Keras, optional system development: Spark
Proactivity, good communication skills, enthusiasm about data, cooperation ability
Login
