Group Members:
Weiyun Jiang, Alexis Yang, Ning Du and Jiajun Wan
Advisors:
Professor Yogananda Isukapalli (University of California, Santa Barbara)
Professor Bruce Schulte (Western Kentucky University)
Statement of Problem
Elephant-Human Conflict (EHC) is one of the major problems in most African and Asian countries. As humans start to utilize natural resources for their own development, the living area of African elephants start to decrease gradually. As the amount of living area for elephants decreases, it becomes more and more frequent and common for elephants to invade human living area for crop-raiding. EHC has caused massive negative economic and moral consequences. A lot of elephants and people died during the crop-raiding. For instance, 100 humans (in some years it may be 300 people) and 40-50 elephants are killed during crop-raiding in India. About 500 people are killed by elephants each year according to the National Geographic Channel documentary Elephant Rage. In addition, millions of dollars have been spent or lost during EHC. For example, in the largest palm oil producing province in Indonesia, Riau, losses due to elephant damage of oil palm plantations and timber estates are estimated to be around US$105 million per year. The crop-raiding affect not only large plantation, but also small ones. Small farmers can lose their livelihood overnight: one adult elephant can eat 200-600 pounds of food per day📷
Forest rangers in Africa spend lots of time herding the elephant away from the villages and crops. At the beginning, they tried to expel these giants via throwing stones, making noise, flashing lights and so on. As time passes, these elephants get used to these actions. A new method of herding elephants are required immediately. Luckily, a group of researchers at Duke University found that the elephants are afraid of the sound of drones. In the experiments conducted by Nathan and Hahn [1], they performed 51 trials to study the effect of drones on elephants. As a result, in every trial, the elephants got scared by the drone at a distance of 50 meters. In fact, the rangers in Africa is controlling these drones manually to herd elephants. The whole process of herding elephants can be achieved autonomously using computer vision, GPS-VHF tags, object detection and so on.
Objectives
Our main goal can be divided into two parts: a GPS-VHF tag for tracking the male adult elephants and a thermal camera and computer vision based drone.
GPS-VHF Tracking Tags
To begin with, we need to construct GPS-VHF tags for tracking the male adult elephants in a given area because the male adults elephants are generally the culprits. VHF(Very High Frequency) serves to estimate the range of the elephants. As elephants come closer to the village, the intensity of the received signal increases accordingly. We can set a buffer zone of approximately 1 miles. Once any elephant invaded the buffer zone, we would activate the GPS module on the tracking tags, sending out the real-time locations. Then, the drone will take off and fly towards the GPS location.
Thermal Camera And Computer Vision Based Drones
Thermal camera works properly during both daytime and nighttime. We plan to implement computer vision algorithms on thermal cameras to distinguish elephants from other animals, such as giraffes, elands and so on. We also plan to estimate the number of elephants. If there were a large number of elephants by any chance, more drones would take off to complete the mission. If the number of available drones are limited due to battery charging or budget, the drones will use a smart scheduling algorithm to prioritize the selection of targets.
Hardware
The project requires intensive onboard computing, so Nvidia Jetson will be used for most of the image processing tasks. Jetson is a SOC development board that contains CPU, GPU, DRAM, peripheral interfaces, and so on. The project also requires a flight controller such as PixHawk 2.1, a thermal camera, distance sensors, GPS, and RF transceiver modules. The specific list is list below:
FLIR Lepton 3.5 160x120
Adafruit Ultimate GPS Breakout
SIM800L
The DJI N3 Flight Controller
The Auvidea J120 carrier board
Algorithm
There are two suitable and fast object detection algorithms for this project, YOLO and SSD. Compared with R-CNN, a popular and well-known object detection algorithm, YOLO and SSD require less computational resources and computational time. Between YOLO and SSD, the former is even faster, but the accuracy of YOLO might decrease. To ensure the video can be processed in real-time, we skip frames for the detector and run a tracking algorithm for every frame instead. The tracking algorithm can be a correlation-based tracker like KCF, or traditional feature tracker such as optical flow. The detector will update the tracker when it is enabled. Additional control and motor coordination algorithms will also be required to navigate the drone correctly. If a multi-drone system is used, extra control software will be needed as well.
References
[1] Hahn, Nathan, et al. "Unmanned aerial vehicles mitigate human–elephant conflict on the borders of Tanzanian Parks: a case study." Oryx 51.3 (2017): 513-516.