To simulate event cameras, which can capture brightness changes at microsecond-scale speeds, conventional methods render numerous frames from 3D scenes, resulting in high computational costs. Researchers from Japan developed an efficient method using physically based path tracing to precisely determine event timings. The efficient implementation using GPUs reduces computation time to as little as one-third of a naïve implementation, enabling more efficient training-data generation for event camera applications, including healthcare monitoring, autonomous driving, and robotics.
From our smartphones to advanced industrial systems, cameras have become an integral part of modern life. Conventional cameras capture scenes as a sequence of frames at fixed time intervals. In contrast, event cameras detect changes in brightness at each pixel as they occur and record them asynchronously as "events." This allows event cameras to capture rapid visual changes within microseconds, while consuming little power and working across a wide range of lighting conditions. These properties make them useful for tracking fast-moving objects, 3D scanning, and robot navigation.
However, because event cameras are not yet widely available, collecting large and diverse event datasets remains difficult, which hinders the development of event-based vision systems, such as autonomous driving. Researchers have therefore developed methods to generate event camera data from 3D computer graphics or conventional RGB video frames. However, these methods require rendering a substantial number of frames to track the microsecond-level visual changes detected by event cameras, making them computationally expensive.
To address this computational cost, the researchers developed a more efficient method for simulating event cameras using physically based path tracing and adaptive temporal search.
The study was led by Associate Professor Hiroyuki Kubo from the Graduate School of Informatics, Chiba University , Japan, and included Mr. Yuichiro Manabe from the Graduate School of Science and Engineering, Chiba University; Dr. Tatsuya Yatagawa from the School of Social Data Science, Hitotsubashi University, Japan; and Dr. Shigeo Morishima from the Faculty of Science and Engineering, Waseda University, Japan.
This paper was published online in the journal IEEE Transactions on Visualization and Computer Graphics on September 17, 2026.
Although path tracing itself is computationally expensive, the proposed simulator avoids rendering a densely sampled sequence of frames. The researchers combined path tracing with a bisection-based search method that determines when a brightness change to trigger an event is observed. They progressively narrowed down the time at which a brightness change reaches the threshold needed to generate an event. Their method could make it easier to generate realistic event camera data for developing and testing vision systems.
"Our simulator allows researchers and engineers to generate physically accurate event streams from virtual 3D scenes—including rare or hazardous scenarios such as nighttime traffic accidents or fast-moving obstacles—and to prototype and validate their algorithms in simulation before deploying them on real hardware," says Dr. Kubo.
The method first uses a bisection-based search to determine when a brightness change crosses the threshold required to generate an event. It repeatedly divides the time between two keyframes into smaller intervals, progressively narrowing down the time when the brightness change reaches this threshold. This allows the simulator to detect events at high temporal resolution without rendering a large number of frames.
However, this process still requires many path-tracing calculations, as the search is repeated at each step. To reduce this workload, the researchers introduced a branch-pruning technique based on statistical hypothesis testing. The technique identifies time intervals where an event is unlikely to occur and stops the search in those intervals, avoiding unnecessary calculations. The implementation further reduces computational cost through GPU acceleration and stream compaction, which allows it to process only the pixels that still require evaluation.
The researchers tested their method using three virtual scenes containing rapidly moving objects: a dynamic Cornell box, bouncing balls, and a fireplace. The scenes contained 20 keyframes over 0.1 seconds, and the researchers also rendered 20,480 frames to generate a high-temporal-resolution reference for comparison.
The results showed that the proposed method generated realistic event streams more efficiently than existing frame-based methods and a path-tracing method using bisection alone. By combining statistical hypothesis testing with GPU-based optimization, the researchers reduced computation time to as little as one-third of that required by the bisection method alone.
By reducing the computational cost of generating realistic event streams, the proposed method could help researchers create large-scale training and benchmark datasets for event camera-based vision systems. These datasets could support the development of applications such as autonomous driving, robotics, and high-speed industrial inspection.
"These research findings are expected to accelerate research on event cameras, for which real event cameras and large-scale datasets remain difficult to access, and to contribute to the generation of training data for artificial intelligence applications such as autonomous driving and robotics," says Dr. Kubo.
To see more news from Chiba University, click here .
About Associate Professor Hiroyuki Kubo from Chiba University, Japan
Dr. Hiroyuki Kubo is an Associate Professor at the Graduate School of Informatics, Chiba University, where he researches image information processing, computational photography, computer vision, and computer graphics. His work focuses on developing imaging technologies that can visualize objects and phenomena invisible to the human eye by measuring and analyzing how light interacts with objects. His research also spans computer graphics applications, including animated video production, stage production, and texture reproduction. He is a member of the Institute of Electrical and Electronics Engineers (IEEE) and the Association for Computing Machinery (ACM).
Funding: This study is jointly supported by Grants-in-Aid of the Japan Society for the Promotion of Science (JSPS KAKENHI JP24K02953, JP25K21810) and the Fusion Oriented Research for disruptive Science and Technology program of Japan Science and Technology Agency (JST FOREST JPMJFR206I).
Reference:
Title of original paper: Path-Tracing-Based Event Camera Simulation via Event-Adaptive Time Refinement
Authors: Yuichiro Manabe1, Tatsuya Yatagawa2, Shigeo Morishima3, and Hiroyuki Kubo1
Affiliations:
1Graduate School of Informatics, Chiba University, Japan
2School of Social Data Science, Hitotsubashi University, Japan
3Faculty of Science and Engineering, Waseda University, Japan
Journal: IEEE Transactions on Visualization and Computer Graphics
DOI: https://doi.org/10.1109/TVCG.2026.3726999
Faculty and Graduate School of Social Data Science Administration Office,
Hitotsubashi University
Address: 2-1 Naka, Kunitachi, Tokyo 186-8601 Japan
