Satellite, Aerial, and Ground Sensor Fusion Experiment for Management of Elephants and Rhinos and Poaching Prevention

View Video Presentation: https://doi.org/10.2514/6.2022-1270.vid Poaching of elephants and rhinos for their tusks and horns has devastated their populations in the wild. Long-term strategies such as poverty alleviation and education in areas adjacent to animal habitats have great merit, however, it is imperative to also protect these animals in the short term. Embry-Riddle Aeronautical University and California State University are designing field experiments to evaluate the potential benefits of sensor fusion in contrast to employing a singular sensor modality, for detecting, tracking, and localizing elephants, rhinos, and people. The proposed sensor network architecture consists of three primary elements: perimeter, range mesh, and aerial. The perimeter element would consist of sensors capable of detecting human or vehicle presence and possible tampering or breaches of the perimeter boundary. The range mesh element would employ GPS collars on select animals in a herd, satellite images, post-mounted acoustic sensors and camera traps, and seismic sensors. The fusion of computer-vision-enabled camera traps, sensors, machine learning applied to satellite images, and GPS collar data would serve to detect, track, and localize animals within the vast game reserves. The final aerial element would be a UAV fitted with high resolution visible and infrared cameras and LIDAR that can be launched on cues received from the perimeter or range mesh elements if a breach or unexpected human presence is detected, or a potential animal-human conflict is predicted. The goal is to test whether the fusion of all these sensing modalities would improve detection and lead to a significantly reduced response time from game reserve personnel in support of conflict resolution. Preliminary field tests in northern Arizona and northern California focused on perimeter detection by using fence mounted accelerometers, camera traps, geophones, and satellite images. The signal processing algorithm for the accelerometers successfully detects shaking, kicking, and climbing events, and Deep Learning applied to the camera trap images detects cattle. Bench testing of a geophone and a detection algorithm picks up human footsteps and will soon be deployed for field testing. Satellite image analysis using Deep Learning shows promise, but very limited image sets of cattle, vehicles, and elephants hampers network performance. Future work will expand the sensors under consideration, design sensor fusion for real-time monitoring, and conduct field tests on a game reserve in South Africa for scalability of the approach.