At Vidhance, Master's thesis projects play a crucial role in driving our exploratory innovation. Each year, we welcome a select group of students to contribute to cutting-edge research in areas that align with our vision for future advancements. Many of our past thesis students have gone on to join our team, continuing their journey with us after completing their projects.

Below are the available thesis topics for Spring 2026. 

1. Effect of annotation noise on model performance
2. Jello effect correction
3. Dynamic multi camera video stitching
4. Video Dehazing´
5. Machine learning vision based relative positioning

1. Effect of Annotation Noise on Model Performance
The objective of this thesis is to systematically investigate the impact of annotation noise on the performance of machine learning models. The project involves designing experiments to introduce controlled levels of noise into annotations within our own datasets. By evaluating model performance under these conditions, the research aims to uncover the effects of varying noise levels on model accuracy and robustness. This study will contribute valuable insights into the challenges posed by noisy labels, ultimately informing best practices in data annotation and model training to enhance reliability in real-world applications.

2. Removing “Jello” Artifacts in Vibrating Cameras
The objective of this thesis is to develop methods for removing the "jello" artifacts that appear in frames captured by cameras in vibrating environments. Camera sensors typically use a rolling shutter to capture individual frames, resulting in a time difference between the top and bottom of the frame. When the camera is vibrating, this produces images that are distorted in a wavy/sinusoidal manner. These artifacts can render footage unusable, making it difficult to extract stable features or produce viewable content. This project involves investigating methods to quantify this distortion (through image analysis or other sensor data) and developing techniques to remove it using classical or ML-based approaches. Solutions to this problem have broad applications in cameras used in real-world environments, particularly for drones.

3. Dynamic multi-camera video stitching
The objective of this thesis is to develop a real-time algorithm for dynamic multi-camera video stitching. This project involves designing and implementing a system capable of processing simultaneous video streams and merging them into a single, coherent panoramic view. Such a capability addresses the significant limitation of restricted field of view found in individual camera feeds, which currently hinders the situational awareness of modern autonomous and remotely piloted systems. By creating a unified, wide-angle perspective, this research will contribute to more efficient monitoring and control, significantly enhancing the environmental perception and operational effectiveness of these advanced systems.

4. Video Dehazing
This thesis project aims to develop a real-time, software-only solution for video dehazing that enhances visual clarity under challenging conditions such as haze, smoke, fog, or dust. Visual degradation from such environmental factors poses significant challenges in critical applications like search and rescue, military operations, autonomous navigation, and firefighting, where visibility is essential. Existing solutions often require specialized hardware, such as thermal cameras or LiDAR, which can be expensive and impractical for lightweight or cost-sensitive systems. This project focuses on designing an efficient video dehazing algorithm that operates in real-time on standard hardware while maintaining temporal coherence across frames. The goal is to improve the usability of conventional camera systems in harsh visual environments by delivering enhanced scene visibility without relying on external sensors.

5. Machine learning vision based relative positioning
The objective of this thesis is to develop and evaluate a relative positioning system using camera vision and machine learning. In a GNSS-denied situation, relative positioning is a foundational building block for estimating the 3D environment. Furthermore, vision based relative positioning can also provide higher accuracy than basic consumer GNSS. Traditional algorithms can struggle in certain scenarios, hence evaluating a ML based system is of interest. At Vidhance we have our own dataset suitable for development and evaluation.

To apply, please click the "apply for this job" button, specifying your preferred topic(s) in your application. Don't forget to include your transcript of records.