Wildlife Conservation: Anti-Poaching Intelligence - Trail Tracker

Download PDF

Abstract

Poaching in today’s world is one of the most significant threats to wildlife. Poachers use different methods to capture animals. Many commercial poachers use military-grade weapons along with arrows and spears to hunt wildlife. Sometimes, objects called snares (a set of wires tied to trees configured to capture any animal by their leg or neck that gets into it) are also implemented. Poachers also manipulate the animal into large nets, known as trap net, pitfall traps (a vast pit dug in the ground that is layered with leaves and plants) or baits. In this paper, we suggest a new solution that operates in real-time to pursue the cause of wildlife conservation by preventing the poaching of any species of animals – endangered or non-endangered by profit-hungry poachers with the help of Artificial Intelligence(AI) and Internet of Things(IoT). In comparison to previous methods in the same domain, it presents an alternate approach, in the form of a monitoring system that can track poaching activity and predict poachers’ behaviour and alert forest authorities for any suspicious crime.

Keywords: ​Artificial Intelligence, Internet of Things, Anti-Poaching, Wildlife Conservation, Applied Machine Learning

Click to get a unique essay

Our writers can write you a new plagiarism-free essay on any topic

1. Introduction

Poaching is the plunder of wildlife organs. Trade of ivory, animal skin and bones are few of well-known resources acquired by hunting the animals in exchange of money[1]. For example, the illegal ivory trade in Africa causes the continent to lose 100 elephants every day. The problem has rapidly increased from the 1990s to date particularly on elephants for example in 2011 the total world population of elephants was 423,000 down from 1.3million in 1979. In 1994 Tanzania had a population of 355,000 but this number dropped promptly to 180,000 in 1999 and less than that number in 2011. According to data from the Tanzania Wildlife Research Institute, Tanzania is losing 30 elephants daily to poaching[2]. The global tiger population has dropped over 95% from the start of the 1900s and has resulted in three out of nine species to be brought on the brink of extinction. At this rate, these creatures will be endangered and will eventually face extinction. These activities are continuously increasing in the world due to the decrease in the population of the animals and increase in the number of the rarity of some species of animals. In order to protect the wildlife and their habitats from poaching and illegal trade various wildlife conservation agencies are established across the globe. Many rangers are stationed to patrol throughout the conservation areas. Wildlife poaching has negative side-effects that affect local communities, wildlife populations, and the environment. It is a crime fueled by a lucrative black market trade of animal parts. The animal parts are sold as novelty items and are sold for their “medicinal” properties. Chinese use ivory for making arts and utilities. Americans use ivory to make gun and knife handles and as decorative details on these weapons.[3]

Environmental groups, animal rights groups, government agencies, and even the ​Duke of Cambridge are calling for an end to wildlife poaching. The ​United States Fish and Wildlife Service (USFWS), ​The World Wildlife Fund for Nature (WWF), and ​The International Anti-Poaching Foundation (IAPF) are leading international efforts to end wildlife poaching. Kenya’s government delivered a powerful message against elephant poaching and the illegal ivory trade on 30 April by burning 105 tonnes of ivory, worth up to US$220 million.[4]

In order to terminate poaching activities that threaten the natural fauna, sophisticated technology needs to be applied as a better solution over the current anti-poaching efforts. This paper presents a new system for anti-poaching involving image sensor (i.e. RGB cameras) and a remote computing unit (i.e. Raspberry Pi) along with Vision Processing Unit(VPU) like Intel Movidius Neural Compute Stick 2 to process complex machine learning computation on the unit itself and a central computer system for data transmission. With this system, a properly selected machine learning algorithm is implemented on the device, if any suspicious activity is observed three actions will take place. The first action is the computing unit will send a notification to the central computer system, the second action is through internet medium a text notification will be sent to wildlife conservation officers alerting them about illegal events in the area and the third action is the frames will also be sent to the central computer system to be transmitted to the wildlife conservation officers and hence providing them with the live stream of the events.

2. Current Scenario

Each protected wildlife reserve forest has their own wildlife conservation authorities wherein managers and officers of authority recruit field ranger staff for patrolling within the forest region. The recruitment and training is done on the basis of:

  1. Analyse current poaching activity and future trends, both locally and regionally;
  2. Analyse the protected area’s specific issues, i.e. size, terrain, access, mobility, avenues of approach, population centres, targeted species, etc.;
  3. Evaluate current protection plans and capabilities; and
  4. Determine the number of additional staff required (if necessary) and necessary skill sets.[5]

The task of these field rangers is to patrol the forest areas to identify and remove any traps within the forest terrain and to register any poaching case occurred within the forest. Rangers record their findings, including animal signs and poaching activity signs, e.g., snares placed by poachers during the patrol, and therefore one can analyze these records to get insights into the poaching patterns.

3. Related Work

A Sensor Based Anti-Poaching System in Tanzania National Parks: Jamali Firmat Banzi “It is not easy to capture these dangerous animal like elephant and attach with sensors and hence need trained personnel. Also the use of batteries brings about many problems such as pollution and extra radiation. Moreover each battery needs to be changed periodically but capturing animals for this purpose is not easy. Furthermore animals may destroy the sensors due to annoyance.”

4 Motivation

With the introduction of AI and IoT, today real-time prediction and real-time inter-communications, as well as intra-communications, can be achieved. Traditional approaches to stop animal poaching included patrolling forest rangers along a specific path at particular intervals to confiscate snares, arrest poachers, and make other observations. In due time AI algorithms were developed to predict the poacher activity within spatial and temporal dimensions using historical information. Another method used involved collaring animals with wireless sensors so as to track them and report poaching activity around them. However, in order to predict poaching activity, it is important to provide precise historical information which should follow a particular pattern to draw proper correlation to poacher’s behavior. Also collaring may affect animal behavior and cause injury to animals or devices alongside the cost involved in designing such an expensive device. An alternative approach can be taken to prevent poaching activity which involves placing an infrared and surveillance camera at specific strategic location within the forest so that it may provide a proper vantage point to cover maximum area and detect activities within its range attached to smart computing device which will perform the task of detecting the activity and report the authorities if activity falls under suspicious category in real-time.

Figure 1: Sample device that can be used in the system

Figure 2: Proposed Infrastructure of system

5. Proposed Infrastructure of The System

Camera Module: The best method proposed in this paper is the use of access points as this method is cost effective. Access points are used to collect data from sensors attached to trees and compute them on the spot. This method doesn’t have any involvement of animals which makes it much simpler than other methods taken. As the camera is attached directly to the computing device the transfer of data becomes much easier and reduces the need of communication modules. Multiple camera modules are used to cover the entire forest area, and each of the modules is distinguished by the camera id given to each of the camera modules to specify in which area the poaching activity is taking place.

Computing Module: ​The Computing Module receives the data(video frames) from the camera module and estimates the position of the human to detect whether the person is in a poaching state or not. It comprises a Raspberry Pi Version 3 Toolkit and an Intel Movidius Neural Compute Stick 2 which is a Vision Processing Unit that helps in accelerating the rate at which the model performs computations on real time data. After the processing is complete the system raises an alert if any activity classified as suspicious is detected and alerts the forest rangers, else the system continues to monitor the area.

Server: ​The Server gathers all the information from the different camera modules to understand the current situation in the forest and alert the rangers.

Figure 3: Proposed flowchart of the system

6. Proposed Methodology

The system functions in a stepwise manner as follows; The camera module sends the input video frames to the computing module (Raspberry Pi and Vision Processing Unit). The motion sequences that are converted to images are then passed through an image classifier to train(from scratch and retraining) and predict. A variety of Image classifiers used are SqueezeNet, AlexNet, Inception, PoseNet, ResNet, VGG and found that PoseNet yields the highest efficiency by retraining on their dataset. They used OpenPose library to preprocess and extract the skeletal points from the video. Several variations in the representations of the two-dimensional images have been experimented upon. Points highlighting joints in the skeleton are extracted from the RGB video and then converted to a single two-dimensional image which are classified using image classifier. where the video gets divided into the frames so that the system can detect the actions performed by the humans which are detected in the frames that are sent by the camera unit. The action can be classified among various types of pre-classified sets of actions performed by the poachers like aiming a gun, crawling on the ground, making a and much more. The system is trained using the machine learning concept and can detect the poachers in live video feed which allows a live-action recognition of the poachers. This method provides a better solution than the rangers personally patrolling the forest area. The camera modules are attached to the trees which makes it easy to set up the system. The software uses Python, Deep learning libraries like Tensorflow and Keras, Server technologies like Laravel, Firebase and Socket.io, Raspberry OS(Raspbian) , OpenCV and Intel’s OpenVino Toolkit for model optimization.

7. Discussion

The proposed system takes the trees in the forest as a background for deployment of the present modules. The system provides various advantages and can be easily adopted to the current system of poaching in progress. It also gives details about the presence of any human in the forest region. Along with this the disadvantages are also provided in the system, the installation of the camera module on the trees is a challenge as the regions are not specific to the animals but rather the areas which are to be monitored only. To cover a large area many such modules would have to be installed in the forest and if any module falls down or gets tampered by the poachers or animals that region would lose the coverage.

8. Conclusion And Future Work

This research paper presents the trees as a static sensor which can be converted to mobile sensors by applying them to drones. The processing can be done on the backend servers for faster computing and the camera module can solely focus on gathering the input ​video frame data. It also effectively presents an alternate solution to fighting poaching in the forest regions through a method that is very cost effective and does not harm any animals during the process of installation as well as while being under operation. The system can be used to detect thefts in shops and any kidnapping cases can be easily detected using this system. It can also be extended to be applied in the military security domain. New sensors can also be added to the system to improve its processing power. This process can reduce the economic costs incurred by forest authorities by a considerable margin. The proposed system will help the policy makers to take informed and well thought decisions, keeping in mind the demography of animals within a forest reserve, areas prone to greater poaching activities as well as areas with poor reachability within the forest.

The scope of this project can be further extended to :

  • Predictive Analysis for determining vulnerable areas within a forest and analysing poaching patterns.
  • Surveillance over flocks of farm animals.
  • Collaborate with Drones for efficient tracking of poachers.

​9. References

  1. Gurumurthy, S., Yu, L., Zhang, C., Jin, Y., Li, W., Zhang, X., & Fang, F. (2018). Exploiting Data and Human Knowledge for Predicting Wildlife Poaching. ​Proceedings of the 1st ACM SIGCAS Conference on Computing and Sustainable Societies (COMPASS) – COMPASS ’18​. https://doi.org/10.1145/3209811.3209879
  2. Massawe, E. A., Kisangiri, M., Kaijage, S., & Seshaiyer, P. (2017). An Intelligent Real-Time Wireless Sensor Network Tracking System for Monitoring Rhinos and Elephants in Tanzania National Parks : A Review. ​International Journal of Advanced Smart Sensor Network Systems​, ​7​(4), 1-11. https://doi.org/10.5121/ijassn.2017.7401
  3. Tanapa Newsletter July-September 2013 retrieved from http://www.tanzaniaparks.com/,available March 3rd 2014
  4. Duan Biggs. 2016. Elephant poaching: Track the impact of Kenya’s ivory burn. Nature 534, 7606 (2016), 179.
  5. Singh, B., Marks, T. K., Jones, M., Tuzel, O., & Shao, M. (2016). A Multi-stream Bi-directional Recurrent Neural Network for Fine-Grained Action Detection. ​2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)​. https://doi.org/10.1109/cvpr.2016.216
  6. Banzi, Jamal. (2014). A Sensor Based Anti-Poaching System in Tanzania National Parks. International Journal of Scientific & Technology Research. Volume 4.
  7. Bondi, E., Fang, F., Hamilton, M., Kar, D., Dmello, D., Choi, J., Hannaford, R., Iyer, A., Joppa, L., Tambe, M., & Nevatia, R. (2018). SPOT Poachers in Action: Augmenting Conservation Drones With Automatic Detection in Near Real Time. ​AAAI​.

image

We use cookies to give you the best experience possible. By continuing we’ll assume you board with our cookie policy.