Exploiting Lightweight Statistical Learning for Event-Based Vision Processing

Abstract

This paper presents a lightweight statistical learning framework potentially suitable for low-cost event-based vision systems, where visual information is captured by a dynamic vision sensor (DVS) and represented as an asynchronous stream of pixel addresses (events) indicating a relative intensity change on those locations. A simple random ferns classifier based on randomly selected patch-based binary features is employed to categorize pixel event flows. Our experimental results demonstrate that compared to existing event-based processing algorithms, such as spiking convolutional neural networks (SCNNs) and the state-of-the-art bag-of-events (BoE)-based statistical algorithms, our framework excels in high processing speed (2× faster than the BoE statistical methods and >100× faster than previous SCNNs in training speed) with extremely simple online learning process, and achieves state-of-the-art classification accuracy on four popular address-event representation data sets: MNIST-DVS, Poker-DVS, Posture-DVS, and CIFAR10-DVS. Hardware estimation shows that our algorithm will be preferable for low-cost embedded system implementations.

A High-Speed Low-Cost VLSI System Capable of On-Chip Online Learning for Dynamic Vision Sensor Data Classification

“…Our hardware system for AER object classification is based on the lightweight statistical algorithm proposed in [ 30 ]. For integrity of this paper, we will briefly review this algorithm.…”

“…To overcome the above problems, a lightweight AER classification algorithm is proposed in [ 30 ] based on extremely computationally simple binary features and the Random Ferns classifier. According to software simulations, the algorithm can be executed 2–3 orders of magnitude faster than prior works and achieves comparable classification accuracy [ 30 ].…”

“…To overcome the above problems, a lightweight AER classification algorithm is proposed in [ 30 ] based on extremely computationally simple binary features and the Random Ferns classifier. According to software simulations, the algorithm can be executed 2–3 orders of magnitude faster than prior works and achieves comparable classification accuracy [ 30 ]. In particular, the Random Ferns classifier inherently supports fast and simple online learning involving merely a few counting operations [ 31 ], which provides a good opportunity for on-chip learning with simple hardware circuits.…”

“…Therefore, in this paper, we propose a high-speed low-cost VLSI hardware system dedicated to AER object classification tasks with fast on-chip online learning capabilities [ 32 ], based on such lightweight statistical algorithms utilizing the Random Ferns classifier [ 30 ]. The proposed system mainly consists of 3 modules: a motion detector, a bank of binary feature extractors, and a Random Ferns engine.…”

“…The rest of this paper proceeds as follows. Section 2 reviews the lightweight AER classification algorithm [ 30 ] deployed on our VLSI hardware system. The details of the proposed VLSI system architecture and circuits are revealed in Section 3 .…”