Ai for Wildlife

Abstract:

India is home to the highest diversity of wild cat species. Every few years, a census is conducted to count the number of animals in a given area to monitor their population. One such organization we are working with is WII, they do the survey for all the endangered cats in the country ranging from tigers, leopards, snow leopards etc. Each individual cat has a unique pattern like a fingerprint which can uniquely distinguish it from other individuals.

Their pictures are obtained from remotely triggered camera traps which are installed in the remote areas of forest with little to no human activity. The pictures obtained often have ‘blank’ pictures which got triggered simply due to the wind. Lack of awareness and government support causes frequent road

killings, retaliatory killings of these animals. This project aims to address how technology AI can help solve problems in wildlife.

My Role: Apart from inspiration, envisioning the design aspects, technical prototypes of the various solution my role was also to coordinate between all the stakeholders across universities, MSR, national parks, biologists and prepare pitches, decks, high level technical architecture. I functioned as a program manager, designer and providing developer support to train the data, labelling.

Inspiration and Insights:

One of the first steps to build a successful application is to identify the need first. The problems of wildlife conservation are esoteric and easily ignored by the clear majority of innovators in the technical spheres. We seldom understand the needs of conservationists, and conservationists are unaware of technical possibilities available to them. A lot of times solutions are made to problems which never existed in the first place, or they were just built to show off a new technology. A proper user-centric design process was adopted for this project and I started off with building ‘empathy’.

I have always been passionate about animals and nature, have grown up in different parts of India always surrounded by forests and sometimes wild animals as well. 2 years ago, I had started volunteering for Eastern Ghats Wildlife Society.

Empathy:

To understand the current activities of wildlife conservationists, I made a field trip with EGWS to Krishna Sanctuary in search of ‘Fishing cat’ an endangered, medium sized, elusive cat. Very less evidence of the cats exists due to its elusive nature and its operation only at night time. After making a 3-hour long boat journey, we went to the slushy mangrove forests to plant camera traps. Thereafter we left them there for 15 days to get a good dataset. After waiting in anticipation for 15 days we would make the long journey back to collect the camera pictures only to realize a lot of them were blank images, false positives or of some domestic animal. Moreover, if we did catch any pictures by the time we got back the animal had gone too far for us to track again. A lot of times these camera traps would get washed away with the sea as well. At a time to do a survey around 100-200 camera traps are installed and much time goes into sorting the images into useful vs not useful.

Once a good number of pictures are collected of the animal in concern – in this case a fishing cat, the next task is to individually identify the animal. Currently this is done manually by researchers who reply on their eye to make a good judgement. This process is used for any animal survey of patterned species such as tigers, leopards etc.

Lack of awareness in locals, evidence, surveys about the existence of such endangered cats has resulted in many road kills, retaliatory killings, hunting for bush meat all of which are major causes of their decline. Even the government is not focused on these species as they don’t come under the ‘charismatic megafauna’ umbrella.

Hence problems identified:

1. Wait in anticipation to find out only false positives, domestic animals in the camera trap pictures

2. Lot of effort in manually identifying cat pictures – current applications require considerable manual effort to trace out animal pictures, hence less adoption

3. By the time we would reach the camera trap area, animals had gone far away

4. Lack of knowledge of locals, awareness and interest causing road kills, retaliatory killings.

5. No support from government on such initiatives

6. Public unawareness on existence of such cats

7. Encroachment, poaching of animals in protected areas

Personas: we identified two key stakeholders in this effort

1. Wildlife biologist

2. Local villager who lives nearby forests with sufficient internet connectivity in case an animal stray out – which is very often

Problem statement

Brainstorming: To counter the above problems we came up with a number of ideas such as

Enhancing community participation through apps like wildlife tracker – using paw marks sign surveys to know about animals and track them

Better maps for data visualization of terrain

Sending signals through camera traps once an animal is detected

A dashboard for tracking animals through wearable sensors, cameras

Awareness campaign, social media

Much interest was given to how AI can solve problems in wildlife. AI (artificial intelligence) can ‘see, conversate, predict and do work’ for you. By using its capability to ‘see’ we will attempt to educate common man about wildlife by automatically identifying animals through simple images of paw marks, tracks, scat etc. We will reduce the time taken to identify individual animals in camera trap data. Using interactive 3D maps, we will tell the story of the wild cats in an attractive, easy to read, informative manner.

Prototyping and Test

Prototypes were developed for the above as shown below.

The problem statement changed slightly to bigger cats due to more data available for them and new stakeholders such as Wildlife Institute of India and Corbett Tiger reserve whose primary concern were tigers, leopards. We compared each of these ideas against factors such as need, technical complexity, relevance to problem, impact. Much interest was given to AI due to its novelty, efficiency, interest from current government which would help in providing funds and also the huge number of students, population in local engineering schools.

Finally, these ideas were selected to develop: -

1. Use AI to individually detect if an animal is there in camera trap picture or not, if so which animal, and send the signal

2. Individually identify the patterned species.

3. Enabling tourists to use the same app to be able to gather more knowledge about animals through their pictures

4. Using drones mounted with IR cameras to identify poachers, encroachment in protected areas

The incorporation of AI for wildlife would also open this topic to many technical institutions/schools in this region, which would grab their much-needed attention in this field. We switched from fishing cats to tigers, leopards due to them being more endangered and a much better dataset available for them.

Wildlife Biologist Persona: Since the wildlife researchers would mostly be in an enclosed control room, a progressive web app was decided which could run on desktops of any resolutions.

Local Tourists, Villagers: To cater to the local villagers, tourists a mobile app was decided.

Both could be combined in the form of a responsive web app with different interface for wildlife researcher and tourist/local in the same app. The interface is as follows: -

To encourage active engagement and participation a leaderboard was introduced apart from the main features of counting animals for each species.

For wildlife researcher’s data visualization graphs are introduced to quickly visualize the status of biodiversity.

Branding colors of forest used – shades of green and brown to cater to semi-arid forests of India and Africa. Customvision was used to identify the animals.

For the drone’s solution much of the problem I focused was on getting the right kind of drones – long range, fixed wing drones to Corbett Tiger Reserve. Parallelly in collaboration with University of southern California, Microsoft Research, efforts were made to get the backend running and deployed at CTR as a pilot programme.

All Recent Work