Asynchronous visual event-based time-to-contact

Xavier Clady, Charles De Clercq, Sio-Hoï Ieng, Fouzhan Houseini, Marco Randazzo, Lorenzo Natale, Chiara Bartolozzi, Ryad BenosmanFrontiers in Neuroscience 2014 · DOI

Abstract

Reliable and fast environmental sensing is critical for autonomous mobile robotics. Traditional frame-based vision systems are often hampered by low temporal dynamics and high computational costs due to redundant data processing. This paper introduces a novel obstacle avoidance method utilizing an asynchronous event-based time-encoded imaging sensor. By computing Time-To-Contact (TTC) from visual event-based motion flows, the system allows every incoming event to contribute to the computation process, enabling high-speed reactive control with minimal energy consumption. Validation on an indoor mobile robot demonstrates that event-based sensing offers a robust alternative to conventional active sensors like laser range finders.

The Problem with Frames

Conventional computer vision relies on a "frame-based" paradigm, capturing snapshots of the world at fixed intervals (e.g., 30-60 Hz). For fast-moving autonomous platforms, this approach introduces significant latency—critical changes can occur between frames. Furthermore, frame-based systems process every pixel regardless of whether it has changed, leading to massive data redundancy and wasted computational power. While active sensors like LiDAR mitigate some issues, they face challenges with specular reflections and high power requirements.

The Event-Based Paradigm

To overcome these limitations, the researchers utilized an Asynchronous Time-based Image Sensor (ATIS). Unlike standard cameras, each pixel in an ATIS sensor operates autonomously, generating an "event" only when it detects a change in relative brightness. This results in a continuous, sparse stream of data that captures temporal dynamics with microsecond precision.

Comparison of traditional frame-based acquisition (top) and the asynchronous event-based stream (bottom), where events are plotted in a spatio-temporal (X, Y, t) volume.

Methodology

The core of the proposed method involves three primary steps:

  1. Event-Based Visual Motion Flow: Instead of comparing discrete frames, the system treats the event stream as a spatio-temporal surface. It calculates the "normal component" of visual flow by analyzing the local timing of incoming events.
  2. Focus of Expansion (FOE) Estimation: As a robot moves forward, the visual scene expands from a central point known as the FOE. The authors implemented a dynamic probability map where incoming events "vote" for the FOE's location. An exponential decay function ensures the system prioritizes recent events, allowing for real-time tracking.
  3. TTC Calculation: By combining the FOE location with the visual flow, the system calculates $\tau$ (TTC) for specific regions. Because the process is event-driven, the TTC estimate is updated asynchronously with every firing pixel, rather than waiting for a full frame to be captured.

Experimental Validation and Results

The method was tested using a Pioneer 2 robotic platform equipped with an ATIS sensor and a Hokuyo Laser Range Finder (LRF) for ground truth. The robot navigated indoor environments, including textured obstacles and warehouse corridors.

  • Ultra-High Temporal Resolution: The system achieved an update frequency exceeding 16 kHz, far surpassing the capabilities of standard high-speed cameras.
  • Computational Efficiency: Processing each event took approximately 20 μs on a standard i7 processor. In real-world tests, the system required only 0.33 seconds of CPU time to process 1 second of high-speed movement.
  • Accuracy: Comparison with the LRF showed that over 60% of the vision-based TTC estimates had a relative error of less than 20% during stable motion.

Conclusion and Future Work

This event-based approach demonstrates that computation can effectively "follow information," scaling naturally with the complexity and speed of the environment. While the current model assumes constant velocity and requires some scene texture, it provides a foundation for ultra-low-latency reactive robotics. Future research will focus on decoupling rotational movements from translational expansion to improve robustness during complex maneuvers.

Summary updated

Abstract

Reliable and fast sensing of the environment is a fundamental requirement for autonomous mobile robotic platforms. Unfortunately, the frame-based acquisition paradigm at the basis of main stream artificial perceptive systems is limited by low temporal dynamics and redundant data flow, leading to high computational costs. Hence, conventional sensing and relative computation are obviously incompatible with the design of high speed sensor-based reactive control for mobile applications, that pose strict limits on energy consumption and computational load. This paper introduces a fast obstacle avoidance method based on the output of an asynchronous event-based time encoded imaging sensor. The proposed method relies on an event-based Time To Contact (TTC) computation based on visual event-based motion flows. The approach is event-based in the sense that every incoming event adds to the computation process thus allowing fast avoidance responses. The method is validated indoor on a mobile robot, comparing the event-based TTC with a laser range finder TTC, showing that event-based sensing offers new perspectives for mobile robotics sensing.
BibTeX
@inproceedings{clady2014asynchronous,
  author = {Xavier Clady and Charles De Clercq and Sio-Hoï Ieng and Fouzhan Houseini and Marco Randazzo and Lorenzo Natale and Chiara Bartolozzi and Ryad Benosman},
  title = {{Asynchronous visual event-based time-to-contact}},
  booktitle = {Frontiers in Neuroscience},
  year = {2014},
  doi = {10.3389/fnins.2014.00009},
}
Figure