Acoustic, Seismic, and Visual Camera Sensor Fusion Experiments for Large Animal Detection and Tracking with Scalability

Abstract

Prior research efforts to detect and track large animals have relied upon a combination of ground, aerial, and satellite remote sensing. For example, detection of elephants and human activity to help with the prevention of the poaching of elephants and rhinos for their tusks and horns has focused on large scale sensor networks that are multi-domain (ground, aerial, satellite, perimeter fencing). These networks show promise, but data from satellite and aerial domains can be expensive, subject to limitations of weather, terrain, and foliage challenges, whereas interior mesh and perimeter sensors can offer cost and performance advantage, but only if spatial scaling can be achieved. While all sensor domains are likely to provide valuable information for game parks, ranches, and farms that need wide area large animal monitoring, the interior mesh sensor networks can be improved by combining sensing modalities to go beyond common practices like use of camera traps. In this paper, we examine cost effective methods to build multi-modal ground sensors for open range interior mesh use, which combine seismic, acoustic, and visual sensor data to improve large animal detection, and re-detection within scalable mesh networks for tracking. The experimental sensor network is focused on open range interior mesh nodes including nodes in an edge, fog, and cloud hierarchical network to enable scaling to larger areas. The scaling of interior mesh sensors provides improved spatial coverage, combined with high fidelity spectral and temporal resolutions possible with edge computing in a multi-modal sensor assembly described along with methods of sensor and information fusion. The use of multi-modal, scalable mesh sensor networks is compared to multi-domain methods in terms of overall cost and performance. The goal is to evaluate whether the fusion of all these sensing modalities would improve detection and lead to a significantly reduced response time from game reserve, ranch, or farm personnel to resolve issues with large animal health and safety. Preliminary field tests in northern California focused on use of mesh sensor assemblies that combine seismic, acoustic, and visual sensing are described along with preliminary testing and performance results. Future work will focus on evaluating the scaling and efficacy of this design at a game reserve in South Africa.