SERENGETI SELF­IES

Mo­tion-trig­gered cam­eras cap­ture au­then­tic, ‘breath­tak­ing’ views of wildlife

Los Angeles Times - - FRONT PAGE - SEAN GREENE sean.greene@la­times.com

Re­searchers placed 225 cam­eras around Tan­za­nia’s Serengeti Na­tional Park and got some amaz­ing — and of­ten amus­ing — images of wildlife there, in­clud­ing a Thomp­son’s gazelle, above.

Hun­dreds of thou­sands of Serengeti “self­ies” are giv­ing re­searchers a can­did, and of­ten amus­ing, pic­ture of what life is like on the African plain.

Ze­bras munch on dried grass, lion cubs play with their mother, ele­phants go for a stroll, and birds hitch a ride on the back of a wart hog.

The images were cap­tured by 225 “cam­era traps” set up through­out Tan­za­nia’s Serengeti Na­tional Park. The cam­eras took a pic­ture when they de­tected mo­tion nearby. Of­ten, the pho­tos were snapped with the an­i­mals star­ing di­rectly into the lens.

Ecol­o­gist Alexan­dra Swan­son de­ployed the cam­eras in 2010, when she was a grad­u­ate stu­dent at the Uni­ver­sity of Min­nesota and wanted to ob­serve how the Serengeti’s preda­tors in­ter­acted with other species. She spent years driv­ing from cam­era to cam­era, chang­ing their mem­ory cards and bat­ter­ies ev­ery two months.

“When we first started get­ting th­ese pho­tos, they were just breath­tak­ing,” said Swan­son, who is now a re­search fel­low at the Uni­ver­sity of Ox­ford in Eng­land “You’re see­ing th­ese an­i­mals at their most au­then­tic.”

Those an­i­mals in­clude some of the Serengeti’s most fa­mous fauna, such as li­ons, chee­tahs, ba­boons and hip­popota­muses. The cam­eras also spied lesser-seen species such as honey badgers, zo­ril­las and aard­wolves.

Cam­era traps al­low re­searchers to ob­serve wildlife in re­mote lo­ca­tions and to mon­i­tor species as they move across vast ar­eas. The re­sults are of­ten as amus­ing as they are en­light- en­ing.

“You just see things you’d never oth­er­wise see — th­ese an­i­mals mak­ing ridicu­lous faces, peer­ing into the cam­era, run­ning to­ward the cam­era,” Swan­son said.

From a re­search stand­point, the au­to­mat­i­cally trig­gered cam­eras are ap­peal­ing be­cause they’re cheap and not very in­va­sive to wildlife (though ev­ery once in a while, an ele­phant might smash one or a hyena might eat one). How­ever, they pro­duce a mas­sive vol­ume of images, mak­ing it dif­fi­cult to process the re­sults.

“A lot of peo­ple who use cam­era traps are of­ten over­whelmed by the num­ber of pho­tos they have to go through,” Swan­son said.

In three years, Swan­son’s cam­eras amassed 1.2 mil­lion sets of images, which were pub­lished and de­scribed this week in the jour­nal Sci­en­tific Data.

Comb­ing through and an­a­lyz­ing such a large num­ber of images was too much for one per­son. Swan­son ini­tially re­cruited a dozen un­der­grad­u­ates to help, but still they couldn’t keep up.

Swan­son asked fel­low ecol­o­gist Mar­garet Kos­mala, whose back­ground is in com­puter science, if there was a way for a com­puter to an­a­lyze the pho­tos.

“I said no,” said Kos­mala, who is now a post­doc­toral fel­low at Har­vard Uni­ver­sity. “Com­puter vi­sion re­search isn’t ac­tu­ally there yet in terms of iden­ti­fy­ing an­i­mals in pic­tures.”

But then Kos­mala saw the images and was blown away. She thought maybe the re­searchers could en­list vol­un­teers — hun­dreds of them — to help.

So the pair teamed up with Zooni­verse, a web­site that hosts cit­i­zen-science projects, to cre­ate Snap­shot Serengeti.

Since its launch in 2012, 30,000 peo­ple have logged on to Snap­shot Serengeti to help make 10.8 mil­lion clas­si­fi­ca­tions. The vol­un­teers found an­i­mals in more than 300,000 images and iden­ti­fied 40 species.

“We lit­er­ally couldn’t have gone through all of the pic­tures with­out the vol­un­teers,” Kos­mala said. “This is science that couldn’t have hap­pened with­out them.”

Zooni­verse be­gan with an as­tron­omy project called Galaxy Zoo, which asked vol­un­teers to help iden­tify the shapes of gal­ax­ies. The site has since ex­panded to 42 projects across a va­ri­ety of dis­ci­plines.

Any­one can par­tic­i­pate.

“The plat­form is de­signed in a way that whether or not they’re an ex­pert, any­one can make a con­tri­bu­tion,” Swan­son said. “It doesn’t mat­ter if you have no idea what you’re look­ing at.”

Snap­shot Serengeti presents users with a photo and asks them to choose from 54 types of an­i­mals, in­clud­ing birds, rep­tiles, in­sects and even hu­mans, to clas­sify the sub­ject. Some­times there’s noth­ing in the im­age.

The user can nar­row the search by char­ac­ter­is­tic, such as color, pat­tern or horn shape. The site also asks about other de­tails, such as how many an­i­mals there are and what they’re do­ing.

To help ver­ify the vol­un­teers’ iden­ti­fi­ca­tions, the site shows the same im­age to mul­ti­ple peo­ple.

“If 10 peo­ple say it’s a ze­bra, then we know it’s a ze­bra,” Kos­mala said. “If many peo­ple say it’s a ze­bra and one says it’s a gi­raffe, then we still know it’s a ze­bra.”

When Swan­son and Kos­mala com­pared the cit­i­zen sci­en­tists’ re­sults with an ex­pert’s iden­ti­fi­ca­tions, they found their vol­un­teers were cor­rect 97% of the time. That’s im­pres­sive, Swan­son said, es­pe­cially when you con­sider that even ex­perts make mis­takes.

Swan­son said she hoped the images could be used for fur­ther eco­log­i­cal re­search and ed­u­ca­tion. Sci­en­tists can view the images for their own stud­ies on the Serengeti or to com­pare the re­gion’s wildlife with that in other ar­eas.

The project is also con­tribut­ing to com­puter vi­sion re­search by help­ing train com­put­ers to rec­og­nize an­i­mals. That re­quires large sets of images with the sub­jects al­ready iden­ti­fied.

Some im­age sets like this al­ready ex­ist, but most of those pic­tures are well­com­posed and well-lit — two things that are hard to achieve with au­to­mat­i­cally trig­gered cam­eras. Cam­era traps of­ten catch crit­ters when they’re out of fo­cus, only partly in the frame or with other an­i­mals, and com­puter al­go­rithms must be trained to ac­count for all of th­ese fac­tors, re­searchers said.

Snap­shot Serengeti

ZE­BRA

LION

BA­BOON

WA­TER BUF­FALO

A CHEE­TAH lunges at a “cam­era trap,” one of 225 mo­tion-trig­gered cam­eras planted through­out Tan­za­nia’s Serengeti Na­tional Park. In three years, the cam­eras col­lected 1.2 mil­lion sets of images.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.