Denis SHEMONAEV will defend his thesis on May 27th, 2024 at 14:00 in the Amphi JP. DOM from the IMS laboratory on the subject : “Definition of an object localization algorithm based on video streams from multiple cameras and evaluation on an FPGA MPSoC”.

The emergence of Convolutional Neural Networks (CNNs) has revolutionized computer vision, particularly in tasks like multi-object detection and tracking crucial for industries such as defect inspection and inventory management. While cloud computing has eased computational burdens by transferring tasks to powerful servers, it presents challenges in real-time processing and data security. On the other hand, edge computing, a cornerstone of Industry 4.0, advocates for localized computational resources. In this context, FPGA MPSoCs offer flexibility, efficiency and computational power. However, integrating computer vision algorithms onto these platforms is challenging. This thesis presents an object localization algorithm for multiple video streams targeting an FPGA MPSoC. It defines a modular algorithm for processing and fusing multi-stream video data with ease of use and development integration in mind. Experiments were conducted to simplify, implement and evaluate a multiple object tracking algorithm on an FPGA MPSoC. This implementation achieves real-time performance while maintaining a detection and tracking accuracy close to a GPU implementation.

Contact our team

If you have a request or questions about the laboratory, please contact our team.