project contest in academic year 2022/23
Object Detection Using Event Cameras
Students: Gil Monat & Ziv Nissim
Supervisor: Michael Fischer
Vision and Image Sciences Lab (VISL), in cooperation with RAFAEL
In this project a novel algorithm was developed and implemented, whose purpose is the detection of moving objects, in video obtained by event cameras.
An event camera is a new type of sensor that collects visual information in an asynchronous fashion. While regular cameras sample all pixels in one frame with a constant frame rate, event cameras capture changes in the pixel intensity, with no fixed rate of sampling, i.e., every pixel gives an indication when there is a change in its luminance. Such a change is called “Event”. The output of the sensor will be a stream of the events it senses. Due to their unique sensing, event cameras have many advantages over classic cameras, which can be
utilized in many ways.
In this project multiple different moving objects where detected, and classified, while each object is a class. It is assumed that each object moves at a constant speed, at least for short enough time slots, and in a direction that is parallel to the camera plane. The newly developed algorithm showed robustness to noise.
With the event classification, objects can be easily identified. In future work we can use this to track very-fast moving objects, applications may be detection of bullets flight, analyzing propeller rotation and many more.
The above figures show a visualization of the data from an event camera of a scene of balls moving quickly in different directions (left), and the point cloud of pixel events with the objects recognized by our algorithm, tagged in different colors (right).
To the laboratory site click here.
