HSMD: An Object Motion Detection Algorithm Using a Hybrid Spiking Neural Network Architecture

Abstract

The detection of moving objects is a trivial task performed by vertebrate retinas, yet a complex computer vision task. Object-motion-sensitive ganglion cells (OMS-GC) are specialised cells in the retina that sense moving objects. OMS-GC take as input continuous signals and produce spike patterns as output, that are transmitted to the Visual Cortex via the optic nerve. The Hybrid Sensitive Motion Detector (HSMD) algorithm proposed in this work enhances the GSOC dynamic background subtraction (DBS) algorithm with a customised 3-layer spiking neural network (SNN) that outputs spiking responses akin to the OMS-GC. The algorithm was compared against existing background subtraction (BS) approaches, available on the OpenCV library, specifically on the 2012 change detection (CDnet2012) and the 2014 change detection (CDnet2014) benchmark datasets. The results show that the HSMD was ranked overall first among the competing approaches and has performed better than all the other algorithms on four of the categories across all the eight test metrics. Furthermore, the HSMD proposed in this paper is the first to use an SNN to enhance an existing state of the art DBS (GSOC) algorithm and the results demonstrate that the SNN provides near real-time performance in realistic applications.

NeuroHSMD: Neuromorphic Hybrid Spiking Motion Detector

“…Although FPGA devices being normally programmed complex Hardware Description Language HDL tools, HLS tools such as OpenCL can be used to increase the productivity of SNNs design process by providing hardware abstraction which reduces the implementation complexity. The NeuroHSMD reported in this paper, improves the speed of the HSMD [14] without degradation of the background subtraction accuracy using an high-end FPGA device.…”

“…The GSOC algorithm has demonstrated better accuracy on the CDnet2012 and CD-net2014 datasets [58], [59] when compared to other algorithms available on the OpenCV library. Machado et al [14] proposed the HSMD model (see Figure 1) inspired by the object motion functionality exhibited by vertebrate retinas, in which Object Motion Sensitive -Ganglion Cells (OMS-GC) determine the difference between a local patch's motion trajectory and the background [9]. The HSMD uses a 3-layer SNN to enhance the GSOC BS algorithm [58], [60], [14].…”

“…The HSMD algorithm [14] has proven to be very sensitive to object motion events triggered by objects as a direct con-sequence of using an SNN to emulate the basic functionality observed in OMS-GC. The SNN utilised is composed of 4 layers of neurons interconnected on a 1:1 synaptic connectivity.…”

“…The scientific community agrees that the multi-core and multi-processor strategy will shortly meet technological limitations related to the increase of power consumption of these solutions [10]. Machado et al proposed the HSMD algorithm [14] that has proven to be very sensitive to object motion events triggered by objects as a direct consequence of using an SNN to emulate the basic functionality observed in OMS-GC (see Figure 1). The SNN utilised is composed of 4 layers of neurons interconnected on a 1:1 synaptic connectivity.…”

“…The NeuroHSMD is the hardware implementation of the HSMD algorithm [14]. As discussed above, FPGAs offer the desired flexibility to accelerate massively parallel SNN architectures.…”