eleID Press Release

Project eleID^© an Automated Elephant Identification Program.

A Project of the Sri Lanka Wildlife Conservation Society

 

Project Requirement

Elephant conservation and management requires information on elephant abundance, distribution,  movement, feeding and habitat use.   The visual identification of elephants is an effective tool that can be utilized to obtain this information since the alternative methods such as radio-telemetry and GPS tracking are very expensive and time consuming.  The initial objective of the eleID^© project is to develop a visual computer based algorithm for automated individual elephant identification. 

Overview       

Utilizing the Google Earth API researchers and the general public will be able to upload pictures of elephants to the eleID^©  system and the algorithm  will search for and match the images to individual elephants that had been previously  identified.  If it is a previously identified individual then the last observed location and date will be logged into the central database.  Over time the logged information will help to track the temporal and spatial movements of individual elephants.   If previously unidentified then the eleID^© system will provide a new unique identity and log the relevant spatial and temporal details accordingly.   The system will treat "public" and "researcher" inputted data differently so that results are scientifically valid.

Manual Methods    

In manual photo based elephant identification, researchers have to learn to memorize discriminating and distinct characteristics of individual elephants to efficiently identify them.  Unfortunately this system does not allow the general public to participate in such a project.   In addition, the manual method is costly and time consuming.  

Mission

The mission of the eleID^© project is to automate this process and make it “user friendly” using "invisible" state of the art techniques based on cost effective technology, which will be complemented by the field experience we have accumulated on elephant identification.  Once operational the eleID^© system will be opened up for "Citizen Science" and will help to obtain reliable and accurate data which will help to estimate elephant populations, home range, habitat use, density and ranging.

Progress

Several visual computer based algorithms have been developed especially for human face recognition over the last two decades.  Research has also been conducted on developing manual elephant identification techniques.  For our project we have thoroughly studied these methods and technologies and have integrated some of the most applicable techniques and are also using some of them as platforms to develop our own techniques and algorithms to develop the eleID^© system. 

Challenges and Obstacles 

In photo based  identification, the differences in light and darkness (light and shade) can be  highly variable and this poses a big challenge for effective image recognition mainly because the face of a particular elephant can appear differently due to changes in lighting conditions. Sometimes the  changes in the images of the same elephant due to different lighting could be  larger than the difference between individuals. This increases the incidences of  false positives where the same elephant can be identified as two or even several different individual elephants. We hope to solve this issue by considering four basic approaches.  They are the heuristic method, different image comparison method, class based method and model based approaches.  Another challenge that needs to be overcome is that when multiple poses of the same elephant is presented  it effects the performance of the identification system.  To overcome this issue  we are planning to apply  a hybrid approach, i.e.  multiple images are made available during the training period but only one image per elephant will be made  available in the database during recognition.  In addition to make the algorithm more accurate additional input data provided by users such as relative tail length, tail tuft, pigmentation, tusks and tushes (if present), ear folds and shape, unique rips, holes, notches and Geo-spatial location information will be used with dynamic weighing mechanisms.

Benefits and community involvement.

The eleID^© system will facilitate obtaining demographic and behavioral information on  elephants which is crucial for elephant conservation and management.  Moreover the technology that is being developed for the eleID^© system can be easily adapted to help gather information on other endangered species for their conservation and management.  Through its ‘user-friendly’ interface the eleID^© project aims to promote public engagement with Asian elephant research and the online Google Earth based database will provide access to information on all aspects of Asian elephants in general.  The increasing prevalence of consumer electronic devices that can record media, such as mobile phones and digital cameras will allow for easier citizen science data collection by the online database on Asian elephants.  This is the incredible potential of the eleID^© system—to provide a common forum for both scientists and citizens to participate to save the endangered Asian elephant.

Funding and Personnel.

The eleID^© project is partially-funded by the Tides Foundation through Google Geo Challenge Grants and is a sub-project of the Sri Lanka Wildlife Conservation Society's (SLWCS) award winning Saving Elephants by Helping People (SEHP) Project.  It is being implemented by an MSc Student, Mr. Ranga Dabarera under the supervision of Dr. Ranga Rodrigo of the University of Moratuwa, Sri Lanka.   The eleID^© Project is carried out under the management and technical expertise of Chandeep Corea, Operations Director and GIS/IT specialist of the SLWCS with the expertise and guidance of Mr. Ravi Corea, founder President of the SLWCS.  

More details can be found at www.SLWCS.org

 

Handling scale and horizontal pose variations.

The images of elephants uploaded by users will be different in scale and the elephants can have different poses in images. So first the scale, pose and illumination variations should be handled. 

Training Image set

In configuring the system initially a training image set of known individuals will be given. Based on those images best features for classification of elephant faces will be determined.

Mean Image (Intermediate stage)

 

This image is from an intermediate process where it shows the mean image of an elephant face.

Field Image

A typical image obtained from the field for identification.

Input image to the system and Reconstructed image  

This image shows the preprocessed input image (left) obtained from the field and the reconstructed image (right) obtained using a waited sum of classifiers or principle face images. The distance of this reconstructed image to the existing image in the database will determine whether the elephant has been already identified (i.e., when the distance is low) or should be given a new identity(i.e., when the distance is high). 

Process Diagram.

                                                                                                                                                                                                                                                       

                                                                

                                                                                                               

---

        

Feedjit Live Blog Stats