Event-based cameras for sports

Abstract

“Of all the senses, sight must be the most delightful.” After losing her sight and hearing, author Hellen Keller (1880 - 1968) reflected in her book “The Story of My Life” on the value of senses and their perspective on the world. Nowadays, cutting-edge vision technologies, such as event-based cameras, can offer a reflection on what vision we can provide to machines.

Traditional vision sensors capture the world as a synchronized sequence of global shutter frames. This format has been the norm for decades in computer vision. This mature technology suffers from multiple critical limitations, such as low dynamic range and motion blur, among others. To overcome these limitations, high-end costly camera sensors are developed, but their availability is limited, and they often require a complex set-up.

Event-based cameras, on the other hand, provide a promising, cost-effective solution to the traditional computer vision limitations. They rely on a valuable set of properties, namely, a high dynamic range (over 120dB), a low power consumption (< 10mW), a high temporal resolution (>10 000 fps, equivalent temporal precision), a high system bandwidth (1.6 Gbps). Thanks to those promising features, Event-based vision is attracting the attention of the public, academia, and industry, making it an increasingly promising and fast-growing field.

However, these sensors require rethinking traditional computer vision methods and adopting new vision paradigms. Indeed, traditional algorithms, applications, and methods do not apply to the unconventional, sparse, and asynchronous output of event-based vision.

This work introduces a comprehensive overview of event-based technologies and principles. Additionally, it explores the opportunities offered by the event-based vision sensors, testing their contributions to a wide range of industrial application scenarios. Next, this work provides a generic exploratory dataset composed of nombre de sequences applicative and generic sequences. Finally, we study the applicative fields for dynamic vision in sports, focusing on the video frame interpolation method provided in .