A digital system using computer vision and deep-learning technology is being deployed to help locate people missing after the devastating Bhotekoshi river floods and identify bodies that have proved difficult to recognise.

With hundreds of bodies still awaiting identification, a group is turning to FaceTagr. This artificial intelligence system analyses faces in photographs and videos and compares them with images of missing people. The technology can narrow thousands of possible records to five or ten likely matches, which can then be reviewed by families.

“This means families no longer have to keep going from one place to another searching,” Samir Panthi, founder of Nowtrkit, a Houston-based company, told Kantipur. “We are simply trying to use technology to identify possible matches and then hand the information over to Nepal Police or the relevant government agencies for official verification.”

Panthi said the system can match living people with up to 99 percent accuracy, as their faces are usually intact and their eyes are open. Matching becomes less reliable after death, when swelling, injuries and closed eyes can significantly alter a person’s appearance.

Possible matches can be reviewed through the khozi.facetagr.com/MissingPersonRegistry portal, where the software generates a limited set of results, said Panthi.

Panthi, an information-technology entrepreneur who has lived in the United States for nearly three decades, is running the initiative with his former college friends Vijay Gnanadesikan and Elango MeenakshiSundaram, who developed FaceTagr, a facial-comparison technology in India.

They have created a separate portal, khozi.facetagr.com, specifically for Nepal. The system uses facial-recognition technology to compare photographs of missing people against images in a database. Families seeking a missing person must enter the person’s name, age, sex, last known location and distinguishing features, and upload a clear photograph.

About 500 missing people have so far been registered on the FaceTagr system, Panthi said.

“We have been able to identify about 90 bodies,” he said. “The families have already received the identified remains.”

Identifying the victims of the floods has proved slow and difficult.

According to Nepal Police, a total of 1,367 bodies and human remains have so far been recovered from flood-affected areas. Of those, 498 are partial human remains.

Only 102 of the recovered bodies have so far been identified and handed over to their families, police spokesperson Abi Narayan Kafle said. That leaves 1,265 bodies and partial remains still unidentified.

Nepal Police has also created a dedicated subdomain on its website to catalogue and publish information on unidentified bodies recovered in connection with the floods.

As bodies began to decompose, raising concerns about disease and foul odours, 335 bodies were buried on August 29 in a forested area near Devghat in Chitwan, with identification numbers placed above ground in line with International Committee of the Red Cross guidelines. Police said the bodies were buried in a way that would allow them to be exhumed and handed over to families if a future DNA match establishes their identity.

“The bodies are extremely difficult to identify because they have been badly damaged and dismembered,” Kafle said. “In some cases, two different families have even claimed the same body.”

Police have collected 2,570 DNA samples and are processing them, but the procedure is lengthy and can take weeks. Of the total, 1,280 samples were taken from the recovered bodies and 1,290 from relatives of missing people.

Nepal Police has also been receiving DNA profiles from overseas through email. Forensic experts from India, China, Singapore and South Korea are assisting Nepali authorities with the process.

Panthi said FaceTagr could help provide possible identifications while families wait for the much slower DNA process.

People searching through the portal are required to provide their name, their relationship to the missing person and a telephone number so they can be contacted if a possible match is found.

Photographs uploaded to the portal are compared with images of unidentified victims collected from the police database and other publicly available sources. FaceTagr’s deep-learning model analyses facial landmarks and features, including the structure and contours of the face, eyes, nose, mouth and chin, to determine potential matches.

Sushil Singh, who is coordinating the technical side of the initiative in Nepal, said the team is comparing photographs uploaded to the search portal with images of missing people found on social media and elsewhere online, as well as the limited number of photographs of unidentified bodies released publicly by police.

Singh and Panthi said they were conscious of the privacy and data-security concerns.

“Families voluntarily register photographs and information about missing people, while photographs of recovered bodies are taken from sources where Nepal Police has made them publicly available. This is not proprietary data,” Singh said. “Once an identification has been officially confirmed, the information is permanently deleted from the system.”

Panthi said the team would hand over all identification-related reports and supporting documents collected through the portal to the government.

He argued that privacy risks were limited because the system contains data only on a relatively small group of missing people, recovered bodies and their families.

Vijay Gnanadesikan, co-founder and chief executive of FaceTagr, said in a LinkedIn post that the technology had been deployed to help locate people missing after the floods in Nepal.

He said families and friends could register missing people through the portal and submit photographs, which would then be compared against images obtained from the affected areas and other available sources.

He described the initiative as a humanitarian, non-profit effort.

Karvika Thapa, chief executive officer and executive director of Kimbu Tech, also sees potential in using such tools.

She said the Rebuilding Nepal group was separately helping families search for missing people. Her team had developed a prototype that used AI to digitally reconstruct damaged facial features from recovered remains and compare them with photographs taken before the person went missing.

“We built the prototype to make it easier for families to identify their relatives, but it has not been deployed,” Thapa said. “Because non-governmental groups face practical and legal difficulties when dealing directly with sensitive information, I believe the state needs to work with the technology community and create a framework for cooperation.”

Nepal Police acknowledges the potential of such technology but says its use must be weighed against legal and operational constraints.

Spokesperson Kafle said some of the bodies recovered from the Bhotekoshi disaster could not be reliably identified by methods other than DNA testing.

“Digital and photographic technology can be useful for bodies that are relatively intact and have not undergone significant decomposition,” he said. “But for bodies that have been battered by mud, rocks and sand, dismembered or badly decomposed, identification based on photographs is not possible.”

Kafle cited the case of a body found wearing a police uniform that was claimed by three different police families.

“In such circumstances, technology can also be misled,” he said. “The final scientific confirmation has to come from DNA testing.”