Burak Ercan

A research manuscript that has been submitted to the scientific journal "IEEE Transactions on Image Processing."

Abstract: Event-based cameras are becoming increasingly popular for their ability to capture high-speed motion with low latency and high dynamic range. However, generating videos from events remains challenging due to the highly sparse and varying nature of event data. To address this, in this study, we propose HyperE2VID, a dynamic neural network architecture for event-based video reconstruction. Our approach uses hypernetworks and dynamic convolutions to generate per-pixel adaptive filters guided by a context fusion module that combines information from event voxel grids and previously reconstructed intensity images. We also employ a curriculum learning strategy to train the network more robustly. Experimental results demonstrate that HyperE2VID achieves better reconstruction quality with fewer parameters and faster inference time than the state-of-the-art methods.

A research paper that has been accepted to the 4th International Workshop on Event-based Vision at CVPR 2023.

Abstract: Event cameras are a new type of vision sensor that incorporates asynchronous and independent pixels, offering advantages over traditional frame-based cameras, such as high dynamic range and minimal motion blur. However, their output is not easily understandable by humans, making the reconstruction of intensity images from event streams a fundamental task in event-based vision. While recent deep learning-based methods have shown promise in video reconstruction from events, this problem is not completely solved yet. To facilitate comparison between different approaches, standardized evaluation protocols and diverse test datasets are essential. This paper proposes a unified evaluation methodology and introduces an open-source framework called EVREAL to comprehensively benchmark and analyze various event-based video reconstruction methods from the literature. Using EVREAL, we give a detailed analysis of the state-of-the-art methods for event-based video reconstruction and provide valuable insights into the performance of these methods under varying settings, challenging scenarios, and downstream tasks.

An AI and computer vision-based automated remote FOD detection system.

I designed, trained, and tested deep learning-based neural networks for computer vision tasks, such as classification, segmentation, and detection, using Python, Keras, and TensorFlow. I also developed an image processing pipeline with classical computer vision techniques for image registration and foreground/background detection utilizing C++ STL, Boost, and OpenCV. Finally, I wrote Python scripts to create, manage, process, and synthesize large image datasets and automate various tasks, including end-to-end system testing.

An autonomous unmanned ground vehicle.

I researched, designed, developed, and optimized terrain traversability estimation algorithms for this system. I also implemented software written in C++ and based it on the Robot Operating System (ROS), which fuses information from multiple sensory modalities, such as cameras and LiDARs. The software works on NVIDIA Jetson AGX Xavier devices in real time, enabling the navigation of an autonomous ground robot in unstructured outdoor environments. Finally, I implemented a ROS-based software package to visualize a terrain traversability map on the video coming from an onboard camera.

A high-technology air command and control system.

I worked on object (ships) detection algorithms and real-time georeferencing software that uses GNSS/INS and DEM data to calculate the coordinates of detected objects accurately.

A fixed-wing sub-cloud autonomous aircraft that can take off and land vertically.

I developed real-time georeferencing software that uses GNSS/INS and DEM data to accurately calculate the coordinates of targets that are tracked with a camera.

An advanced military electro-optic system to detect, classify, identify, and warn of hostile laser threats aiming at a platform—airborne, naval, or land—in a short time with high sensitivity.

I was the leading software developer of the system for five years while the system was rigorously evaluated with qualification tests. During this time, I developed real-time embedded software in C to process sensory data and handle all the communication and control system requirements. I optimized laser threat detection and recognition algorithms and contributed to many software packages that support the main project, such as simulation and test applications.

An article published in Signal Processing: Image Communication Scientific Journal.

Here we first introduce a large-scale synthetic dataset called Synthetic18K. Then, we demonstrate that the pretraining of simple deep architectures on Synthetic18K for person re-identification, attribute recognition, and fine-tuning on real data leads to significant improvements in prediction performances. I mainly worked on training and validation of deep neural networks on person re-identification and attribute recognition tasks, also contributing to writing the original draft of the article.

A paper published at the 17th International Conference on Computer Vision, Theory and Applications | VISAPP 2022.

Here we propose an end-to-end algorithm pipeline for 3D surround view systems that work on embedded devices in real time. I mainly worked on the color correction part of the pipeline, designing and implementing a local color correction method based on Poisson image editing. I used an algebraic multi-grid solver from AmgX (a GPU accelerated library) to solve the resulting large and sparse system of linear equations. Then, I optimized this algorithm to work on resource-constrained embedded devices in real time. Finally, I contributed to the experimental set up and the paper writing.

A light thermal weapon sight with excellent image quality and high range performance.

I was the system's leading software developer for four years. During this time, the project has evolved from prototypes to thousands of devices produced and sold. I designed and developed embedded software in C++ to cover a wide range of product features, including thermal image acquisition and enhancement, communication and control, built-in tests, on-screen displays, and button controls.

An integrated military electro-optic sensor system that incorporates a thermal and daylight camera, laser range finder, digital magnetic compass, and GPS-based locator.

I was the system's leading software developer for four years, creating embedded software while the system was continuously updated with new features and changing components. I designed the system's software with many features, including thermal image acquisition, enhancement, and target coordinate calculation.

A compact, high-performance vision system designed for tanks and other armored vehicles.

The system provides enhanced maneuvering capability and situational awareness under severe conditions. It uses thermal and daylight sensors, covering a 360° field of view. I was the leading developer of the system's embedded software for two years. During this time, the system was designed and prototyped from scratch.