LuvHarris: A Practical Corner Detector for Event-cameras

Abstract

There have been a number of corner detection methods proposed for event cameras in the last years, since event-driven computer vision has become more accessible. Current state-of-the-art have either unsatisfactory accuracy or real-time performance when considered for practical use; random motion using a live camera in an unconstrained environment. In this paper, we present yet another method to perform corner detection, dubbed look-up event-Harris (luvHarris), that employs the Harris algorithm for high accuracy but manages an improved event throughput. Our method has two major contributions, 1. a novel "threshold ordinal event-surface" that removes certain tuning parameters and is well suited for Harris operations, and 2. an implementation of the Harris algorithm such that the computational load per event is minimised and computational heavy convolutions are performed only 'as-fast-as-possible', i.e. only as computational resources are available. The result is a practical, real-time, and robust corner detector that runs more than 2.6× the speed of current state-of-the-art; a necessity when using high-resolution event-camera in real-time. We explain the considerations taken for the approach, compare the algorithm to current state-of-the-art in terms of computational performance and detection accuracy, and discuss the validity of the proposed approach for event cameras.

PUCK: Parallel Surface and Convolution-kernel Tracking for Event-Based Cameras

“…• the Exponential Reduced Ordinal Surface (EROS) representation update (inspired by [21]); • a detector that initializes the puck position at the beginning of the game;…”

“…It is a key component that enables the decoupling of these processes to allow low-latency, high event-throughput, for complex visual processing algorithms. The EROS is similar to the Threshold Ordinal Surface (TOS) [21] with a small modification to the representation decay method defined in Algorithm 1, where k EROS corresponds to the update region size around each event position. EROS can be used at any given point in time by downstream processing, as a 'grey image', with values between 0 and 255.…”

“…EROS can be used at any given point in time by downstream processing, as a 'grey image', with values between 0 and 255. Similar to the TOS [21], the EROS is filtered with a Gaussian blur before further processing. However, as we only process a ROI for both detection and tracking, we perform the blur only on the selected area.…”

“…A key feature of the PUCK-track algorithm is the decoupling of the event-by-event processing from the relatively heavier computation of the detection of the puck position, as proposed in [21]. As the camera potentially produces > 10 million events per second, any complex processing performed event-by-event does not meet the real-time constraint in highly dynamic robotic tasks.…”

“…In this paper, we investigate event-based tracking for the task of air-hockey, with the focus on investigating its accuracy and real-time latency, towards closed-loop control. To this end, we propose a novel tracking method that preserves the event-camera asynchronicity and maximises the eventthroughput, in order to minimise computational latency, inspired by [21]. In particular, we exploit the paradigm of simultaneous parallel processing of (i) the spatial surface representation, that is updated asynchronously, and (ii) the heavier tracking algorithm, based on convolution kernels that run on the most up-to-date surface representation as fast as possible, to detect and track the air hockey puck.…”