Elsevier

Applied Acoustics

Volume 129, 1 January 2018, Pages 92-103
Applied Acoustics

Improving accuracy of elephant localization using sound probes

https://doi.org/10.1016/j.apacoust.2017.07.007Get rights and content

Abstract

Localization of elephants in the vicinity of villages is an important issue in mitigating human-elephant conflict. This paper proposes an inexpensive, effective and non-invasive framework that employs a sound probe technique with an acoustic sensor network to localize elephants. Incorporation of probes in our sensor network eliminates the requirement to explicitly measure temperature and wind velocity for accurate determination of sound velocity. A sensor network has been built and experiments performed by replaying recorded elephant sounds under three different environmental conditions. The results overall show that the system is capable of providing remarkable accuracy under distinct wind and temperature conditions. An identical experimental set up was used to localize wild elephants in Sri Lanka. Our approach enabled localization of wild elephants at a distance of over 500 m from the sensors to within 30 m, providing adequate time for the villages to take appropriate safety measures.

Introduction

Changes in social and ecological conditions precipitated by human needs are resulting in serious depletions in certain animal populations. Human-elephant conflict (HEC) is one such problem resulting in deaths of members of both species. Humans are clearing large areas of land for food production, increasing pressure on traditional elephant habitats. This results in migration of elephants to villages to fulfil their food requirements and ensuing HEC. In Sri Lanka, HEC has is causing more than 60 human deaths and over 200 killings of elephants, annually [1].

Various strategies have been implemented to mitigate HEC. These include implementing electric fencing systems and using satellite imaging and radio and Global Positioning Systems (GPS) collars to detect the presence of elephants [2].

Electrified wire fences are used to restrain elephants into the forest areas. This is an expensive option due to high installation and maintenance cost. In practice, the long-term success with anti-elephant fences has often fallen well below expectation. This is basically because of deficiencies in meeting the considerable demands of meticulous routine maintenance. In addition, wild elephants have low resistance to these barriers as they have learned to demolish the fences using tree branches and thereby enter into the protected areas [3].

Tagging elephants with radio and GPS collars for understanding the behavioural aspects is presently conducted in many parts of the world. However, this methodology is impractical for real-time monitoring of wild elephants due to the battery power limitations in collar systems and the cost of GPS data retrieval. In addition, capturing elephants and fitting GPS collars to these animals is a complex and dangerous procedure which risks lives of both elephants and personals involved. Therefore, it is in practice even impossible to collar elephants which are deemed problematic.

With the advancements in technology, satellite imaging is also proposed as a methodology for tracking wildlife [4]. However, thick vegetation and environmental factors hinder this sophisticated and expensive technology for real-time monitoring of wild elephants.

Recently, interest has grown in the detection of elephants through their low-frequency calls, commonly referred to as “rumbles” [5]. This technique is a safe, practical and non-intrusive methodology to detect these highly social animals that use rumbles for long distance communication. Once the elephants are detected at a sufficient distance from the village boundaries countermeasures such as the use of firecrackers, broadcasting of noise through loudspeakers, and warnings to the local population can be deployed.

However, the variations in environmental conditions seriously affect the performance of such systems [6], [7]. This is due to the sound speed dependency on several environmental phenomena such as temperature, water vapour mole fraction, atmospheric pressure and CO2 content in the air. Cramer [8] has proposed a closed form equation for the relationship between speed of sound and the above environmental factors through a laboratory experiment. However, the speed of sound mainly depends on temperature and effects of other variables are insignificant. The speed of sound in air can be expressed in terms of temperature as CSoundT=CSound01+T273.15 where CSound0=331.45 ms−1 is sound speed at 0 °C [9]. In addition, wind causes to movement of the propagation medium which effectively alters the propagation sound speed between sound source and the receivers [10].

In [6] the possibility of using an acoustic sensor network to detect and localize elephants has been extensively analysed. Wind and temperature variations are identified as the main cause for the deterioration of the performance of the above approach [6], [7]. For instance, a uniform wind velocity of 20 ms−1 results in about a 100 m (20%) error at a distance of 500 m. An additional temperature variation of 4 °C can result in an increase in the error to over 175 m (35%) [6].

In this paper we propose a technique to correct the errors arising from variations in sound speeds due to environmental effects. We propose a system to implement the algorithms that have been developed and describe the results of experiments, under three different environmental profiles, using a sensor network system that we designed. Then, we implement our system in a village area in Sri Lanka and test it for the real application of wild elephant monitoring. Our contribution is to present an effective solution that significantly improves localization accuracy over a conventional acoustic sensor network, and which does not require explicit wind and temperature sensors to compensate for the environmental effects on sound speed.

The remainder of the paper is organized as follows. In Section 2 we summarize the localization algorithm that will be utilized in our proposed system. In Section 3 the technique used to correct the speed of sound to mitigate temperature and wind velocity effects is described in detail. In Section 4 we describe the corresponding experimental setup for tests we conducted in open field and forest environment. Section 5 outlines the experimental results and discusses them further. Section 6 concludes the paper.

Section snippets

Localization algorithm

We consider an acoustic sensor network with N sensors that listen to elephant vocalizations. In consideration of the long propagation range of infrasonic rumble signals [11], it is assumed that all sensors receive an attenuated and noise corrupted replica of a call generated by an elephant (source). Let the time of arrival measurements at sensor Si be ti where i=0,1,,N-1. The infrasonic acoustic sensors are deployed at predetermined coordinates (xi,yi). It is necessary to locate the source

Average sound speed estimation methodology

Our goal is to estimate the approximate speed of sound that signals propagate with respect to each sensor. In order to build a solution that attains the above sound speeds in a real world environmental scenario we first develop a model assuming uniform environmental conditions. Then the model is extended to real world comparable, non uniform and totally unknown environmental conditions.

To locate the elephants accurately, acoustic sensors and some specific sound (chirp) signal generators are

Open field experiment

The hardware system consisted of sensor units which contain microphones (Audio-Technica AT897) and FM transmitters (Fmuser SDA-01A) and sound probes which include chirping devices (see Fig. 3, Fig. 4). Separate laboratory experiment was conducted to investigate the frequency response of the signal sensor microphones detection system. It was confirmed that system performs well in the range of approximately 10–400 Hz and can capture main signal energy components of low frequency elephant calls as

Open field experiment results

The results of the practical implementation of our proposed methodology in an outdoor environment are depicted in the Fig. 9, Fig. 10, Fig. 11 for three different data collection sets. It can be identified that our proposed technique clearly improves the localization accuracy of the sound source under different environmental conditions. The accuracy improves whilst the source is close to the probes. The reason for this is that, when probes and SE are in close proximity, the probes technique

Conclusion

In this paper we have investigated the feasibility of enhancing the accuracy in elephant localization system by utilizing a sound generating probe technique. An inexpensive hardware platform is implemented to evaluate the proposed methodology in an outdoor environment under equivalent elephant localization scenario. Finally, the system was deployed in a real forest environment to localize wild elephants in Sri Lanka. The outdoor experimental implementation verified the significant improvement

Acknowledgment

This work was supported by the Institute for a Broadband-Enabled Society (IBES) and the Melbourne Sustainable Society Institute (MSSI), The University of Melbourne, Parkville, VIC 3010, Australia. Authors would like to express their gratitude to Dr. M. I. Bandara, Mr. K. Dissanayake, Mr. S. Wijerathne, Mr. P. Wijerathne, Mr. B. M. Dissanayake, the supporting team and the villagers from the Swarnapali Gama for their support during the data collection procedure.

References (15)

  • Elephant conservation...
  • V. Galanti et al.

    The use of GPS radio-collars to track elephants (loxodonta africana) in the tarangire national park (tanzania)

    Hystrix, Ital J Mammal

    (2000)
  • Hoare R. Fencing and other barriers against problem elephants; 2003...
  • Gutro R. Satellite data to track wildlife: elephants in space, 17 January 2005...
  • Wijayakulasooriya JV. Automatic recognition of elephant infrasound calls using formant analysis and hidden markov...
  • C.M. Dissanayake et al.

    Propagation constraints in elephant localization using an acoustic sensor network

  • Garstang M, Fitzjarrald DR, Fristrup K, Brain C. The daily cycle of low-frequency elephant calls and near-surface...
There are more references available in the full text version of this article.

Cited by (9)

  • Cognitive IoT system with intelligence techniques in sustainable computing environment

    2020, Computer Communications
    Citation Excerpt :

    There is currently no public benchmark dataset for the evaluation of elephant detectors. However, we can compare the results of the proposed approach with that of related approaches [15,17,23] and [19]. An objective comparison with the results of [19] is not appropriate, since the author uses a dataset for few minutes of evaluation. [15,17,23]

  • Stochastic Computing Design and Implementation of a Sound Source Localization System

    2023, IEEE Journal on Emerging and Selected Topics in Circuits and Systems
View all citing articles on Scopus
View full text