Projecte matriculat
Títol: Image processing algorithms for 3D Object Detection and Motion Tracking
Director/a: RYZHAKOV, PAVEL
Departament: DECA
Títol: Image processing algorithms for 3D Object Detection and Motion Tracking
Data inici oferta: 26-04-2024 Data finalització oferta: 26-12-2024
Estudis d'assignació del projecte:
GR ENG SIS TELECOMUN- GR ENG SIST AEROESP
- GR ENG TELEMÀTICA
| Tipus: Individual | |
| Lloc de realització: |
|
| Nom del segon director/a (UPC): Dr. Alireza Hashemi | |
| Departament 2n director/a: | |
| Paraules clau: | |
| Data science, neural networks, machine learning | |
| Descripció del contingut i pla d'activitats: | |
| Introduction
As aircraft systems become increasingly autonomous and data-driven, the need for robust object detection and motion tracking algorithms grows exponentially. Detecting and tracking 3D objects accurately is essential for understanding their behavior and optimizing processes. In this research, image processing algorithms will be developed and implemented to achieve robust 3D object detection and reliable smooth motion tracking. This research will delve into a combination of traditional image processing techniques and state-of-the-art machine learning approaches. Objectives 3D Object Detection: Develop algorithms to identify and locate objects within a database of images. Utilize image analysis tools such as edge detection and contour extraction to measure geometrical characteristics of the objects. Explore depth information (if available) to enhance accuracy in 3D space. Motion Tracking: Implement motion-tracking algorithms to follow each object's trajectory over time. Address scenarios where multiple objects are present in the same image. Ensure accurate tracking even in cases of occlusion or overlapping objects. |
|
| Orientació a l'estudiant: | |
| Methodology
1. 3D Object Detection Edge Detection: Apply edge detection techniques (e.g., Sobel, Canny) to enhance object boundaries. Extract edges from grayscale images. Contour Extraction: Identify closed contours corresponding to desired objects Filter out noise and artifacts. Estimate object centroids and bounding boxes. Depth Estimation: Convert 2D intensity gradients to 3D information. 2. Motion Tracking Single-Object Tracking: Implement tracking algorithms for individual objects. Handle occlusions and sudden movements. Multiple-Object Tracking: Assign object IDs across frames. Maintain object trajectories over time. Machine Learning Approaches Pretrained Models: Leverage existing deep learning models pre-trained on large datasets. Fine-tune these models on our specific dataset. Custom Training: Collect labeled data for desired objects. Train a custom neural network for object detection. Evaluate performance metrics (precision, recall, F1-score, '). Expected Outcomes Accurate 3D Object Detection: A robust algorithm capable of identifying desired objects in images. Geometrical analysis of the detected objects. Reliable Motion Tracking: Algorithms that track individual objects and handle occlusions. Trajectories plotted over time. Requirements Basic understanding of computer programming Responsibilities This project will be supervised by Prof. P. Ryzhakov and Dr. A.R. Hashemi, who will guide and oversee the progress of the tasks. Note: The proposed plan is subject to adjustments and refinements based on the availability of resources and consultation with the supervisor. |
|
| Requereix activitats hardware: No | |
| Requereix activitats software: No | |
| Horari d'atenció a estudiants per a l'assignació de projecte: |
|