Article 52336 How Toronto police used controversial facial recognition technology to solve the senseless murder of an innocent man

How Toronto police used controversial facial recognition technology to solve the senseless murder of an innocent man

by
Betsy Powell - Courts Reporter
from on (#52336)
_1_main_investigative_team.jpg

Nnamdi Ogba was visiting a friend from his soccer team when he left a northwest Toronto apartment to pick up some food for his fianci(C)e.

As the 26-year-old electrical engineer approached his vehicle, key fob in hand, two men ran up behind him and, without words or glances exchanged, opened fire. Ogba was hit five times in the back. It was just after 11 p.m. on March 16, 2018. He was left to die in the parking lot.

It didn't take long for Toronto police Homicide Det. Jason Shankaran - the lead investigator on the case - to conclude Ogba was an innocent victim killed by people he didn't know. Shankaran's "gut" also told him that unless he found out where the shooters had come from, "we weren't going to solve it."

Soon, the investigators were gathering and piecing together images taken from at least 200 video surveillance systems, starting with footage from cameras at the murder scene at Scarlett Woods Court. And it was that key evidence, a terabyte of data, that eventually led Shankaran and his team to become the inaugural users of the Toronto Police Service's then-brand-new facial recognition software - a tool they used to help find, arrest and, last month, secure murder convictions for the three men who shot Ogba.

The Ogba case underscores that technology's potential, particularly for cracking such a notoriously difficult-to-solve investigation as an apparent stranger-on-stranger shooting without DNA, fingerprints or eyewitnesses. The case also highlights the facial recognition is an inexact process, one that initially yielded two correct suspect matches - and one that wasn't.

Although the police use of the software wasn't contested at trial, "I think we did a good job at articulating the limits of where it could take us," Shankaran said in an interview. The "potential candidates" the tool find are treated as one would a Crime Stoppers tip or confidential source, he notes.

"Any investigator must use it cautiously, and corroborate any selection and use it to include or exclude people," he said. "It's a powerful tool, and it's not just for accused, it could also locate a witness."

Still, police use of facial recognition in any form has become fiercely controversial.

Facial recognition systems look at digital images and then analyze and compare certain biometric patterns against what it sees in an existing set of pictures, trying to find a match. The software uses complex algorithms to measure variables of a person's face, such as the length and width of the nose, the width of the eye sockets and the shape of the cheekbones.

In the U.S., police have come under fire over the potential for racial bias based on research showing the algorithms make more errors when identifying African American, Asian and Indigenous suspects. Some civil liberties advocates want it banned until it can be properly regulated.

And across the continent, police forces have faced public backlash for using an app from the U.S. company Clearview AI that runs facial recognition searches on billions of photographs, many posted to personal social media pages.

Earlier this year, the Star found that the Clearview AI app had been tested far more widely in Canada than previously known, including by Toronto police, who had been the most prolific user of the app in this country before the service stopped its use. (At the time, Chief Mark Saunders said the app had been "informally" tested by some officers.)

But Shankaran's team didn't use Clearview. Instead, it was one of the first to use new software that only became fully operational for Toronto police officers on March 22, 2018 - less than a week after Ogba's murder.

The tool, which Toronto police bought from NEC Corporation of America for $450,000 in March 2018, compares pictures of unknown people - such as those gleaned from a security camera - against the service's internal database of 1.2 million mug shots.

Regardless, Shankaran isn't interested in discussing the broader controversy over facial recognition. Like Toronto police leadership, Shankaran believes cases like the Ogba murder have proven it to be a valuable and successful investigative tool.

Within hours of the shooting at Scarlett Woods Court, the team had obtained grainy surveillance camera images from nearby Toronto Community Housing cameras. The footage showed two shooters from a distance. They could be seen rushing up to and killing Ogba, then running away. The shooters were unidentifiable, but they were wearing distinctive clothing: One had on a black Puma jacket with red sleeves and a yellow Ferrari logo; the other shooter wore a white hooded sweater with white cords attached to the hood.

The cameras also picked up the two shooters leaving the scene in a small, silver SUV, that an eagle-eyed officer - Det. Const. Jason Brady - determined was a Nissan Rogue.

But that vehicle was stolen, as were its' plates.

Brady, Det. Const Sameer Patil and homicide Det. Paul Worden began the painstaking work of tracking the Rogue post-murder. They obtained more surveillance camera footage from shops and banks lining what later determined was the 4.4-kilometre escape route. Fortunately for police, the driver failed to turn on the headlights - so just the running lights were on - making it easier for officers to follow the SUV along the darkened and near-empty west-end Toronto streets, using other nearby vehicles as "markers."

The trail of footage ended when the SUV turned onto Clearview Heights, a residential street with lowrise apartment buildings - but then came a game-changer: Det. Chris Ruhl spotted the parked Rogue, which the gunmen had, apparently, used to try to cover their tracks. It was seized for a forensic exam.

Police next obtained nearby surveillance-camera footage from the night of the murder. The video showed three men exiting the stolen SUV, two of them walking arm and arm, one doing a little skip and dance along the sidewalk. What's more, about an hour earlier, the video caught the same men disembarking a taxi in front of 85 Clearview Heights and walking to the stolen Rouge.

The police now knew the killers were "somewhere in one of these buildings," Shankaran explained. But which one?

Brady then went "backward."

By tracing cab company phone logs, he determined that about 45 minutes before the killing, three men had taken a Beck taxi from 25 Martha Eaton Way to 85 Clearview Heights where they jumped into the stolen Rogue.

With that key information, officers then seized more camera footage from that evening showing the trio, wearing clothing similar to the men in the grainy images, arriving back to the entrance of 25 Martha Eaton Way around 11:30 p.m. - "which is where we get our first facial images of our suspects," Shankaran explained.

In the past, investigators might have circulated such images internally, or released them at a press conference asking the public to help identify the suspects.

But Shankaran, a soft-spoken detective with experience at downtown police divisions and the sex crimes unit, worried that might put vital evidence in jeopardy.

"Once the pictures go out, you might lose the gun, or the clothing, we knew we had the car, but we still didn't know who the driver was," he said. "I figured, we're close, these guys don't know that we found where they were " they looked so comfortable walking into the building."

So, after consulting with Hank Idsinga, then the head of Shankaran's team and now the homicide squad's boss, the team turned to the service's new facial recognition software. Det.-Const Aleksandra Zlobicki sent the Martha Eaton Way clips to members of the team, who had received FBI training on facial recognition.

Within hours they had a "potential candidate" in Abdullahi Mohamed, now 24 - a man who had been recently released on parole after serving a sentence for shooting up an Edmonton bar. Three hours later, the team also had another match and the name of a convicted drug trafficker, Trevaughan Miller, now 24.

But the quality and resolution of the images and clips of a third unknown male from Martin Eaton Way, his face partially obscured by a New York Yankees toque, was not as clear as the other two. Still, the software came back - erroneously - with a third "potential candidate" from the mug shot database, a man named Kafi Abshir.

Shankaran, who has a philosophy degree from the University of Toronto, said he felt confident about the Mohamed and Miller matches, but police were still a far cry from making any arrests.

"We needed to treat this information as an investigative aid," he said. "It's not a fingerprint or DNA. There's still a need to corroborate the potential candidate as we would any other information gleaned from a witness, or otherwise."

Investigators still needed to find earlier images of Mohamed wearing that distinct red and black Puma jacket. Shankaran described the question: "If we execute search warrants and we don't find the jacket - which we didn't - can we get him wearing that same jacket on a different day?"

So police seized more video, which instead showed Mohamed wearing other articles of clothing from that evening. The search for the Puma jacket also led to more footage of the gunmen and driver after Ogba was killed, including sequences of them horsing around and laughing in apartment corridors and elevators.

While they weren't evidence of criminality, those images bolstered the Crown's case during the trial. It showed them pleased with themselves after executing their plan, and not contrite that they had killed an innocent man, as Mohamed later testified he had felt.

Police, meanwhile, concluded Abshir wasn't the third man.

For starters, he lived in Scarlett Woods Court, which didn't fit with the Shankaran's belief that the murder "seemed to be outside people" coming into a rival complex to send a message to their enemies. What's more, the officers who began surveillance found Abshir was sporting a full beard - just a week after the shooting.

The man in the Yankees toque was clean-shaven.

"A big bushy beard in a span of a few days convinced us this is not our guy," Shankaran said.

Once Abshir was eliminated as a suspect, Shankaran and other offices met with him to explain how he had landed, unknowingly, in their crosshairs. Shankaran said Abshir seemed to appreciate that. Later that year, on Aug. 25, 2018, Abshir himself became a homicide victim, killed in a stabbing near Weston Road and Lawrence Avenue West.

In the end, the getaway driver was identified by taxi records showing that a phone number belonging to Abdirahman Islow, now 29, was used to order the cab that delivered the three men to the stolen Rogue.

As he prepared the case for trial, Shankaran sent an email asking Det. Sgt. Shawn Meaney, of the TPS facial recognition team, to explain how Abshir's face was selected.

"There are many factors that affect the search results. The quality and resolution of a probe (unknown) image is one of the main factors," Meaney wrote back in an email in the spring of 2018. "Other factors include the distance of a subject to the camera, size of the image, the pose, illumination and expression, and the visibility of a person's eyes," he added.

Each search initially returns 200 potential candidates. Facial recognition team members then may do side-by-side comparisons before whittling down the number to the potential candidate, Meaney wrote.

"In the end, the decision made by myself or a member of my team, although based on our training and experience, is totally subjective. This is why any information provided to an investigator is an investigative tool only."

And Shankaran had another question: Did the fact Islow was "poking his tongue out" in his mug shot prevent the recognition system from making a match? A person's pose could have an effect on the comparison and decision-making process, Meaney reiterated.

A jury found Abdullahi Mohamed, Abdirahman Islow and Trevaughan Miller guilty of Nnamdi Ogba's first-degree murder on March 13 - not long before the COVID-19 pandemic partially shutdown Ontario's criminal justice system and has, for now, suspended all jury trials. Justice Robert Goldstein is scheduled to deliver their automatic life sentences via teleconference on April 21.

Even the seasoned defence lawyers involved in the case acknowledged the visual evidence proving their clients' involvement was overwhelming - though they disagreed on whether the men intended to kill Ogba, arguing Mohamed fired a warning shot aimed at a drug dealer who owed him money, and that Miller "panicked" and fired a volley of shots, unintentionally killing Ogba. The Crown said this was a fabrication, that the three were street gangsters who went to rival Scarlett Wood Court, guns blazing, to intimidate the entire community.

Either way, with the video evidence, there was never going to be a "no face no case," as the slang saying goes.

"Their behaviour was on camera from beginning to end," said defence lawyer John Struthers, who represented Islow, the driver of the getaway car who denied knowing anything about a plot to kill a random target.

"It never was an identification case ... there's like HD Japanese television quality images ... better than passport photos," said Struthers, who is president of the Criminal Lawyers' Association and has been representing scores of accused killers for decades.

Added Greg Leslie, who represented Miller: "This was good, old fashioned police work."

Despite the limitations, Shankaran believes the TPS in-house facial recognition system is a valuable investigative tool that, in this case, helped ensure the perpetrators of an innocent man's execution were held to account.

"It allowed us to quickly jump ahead using it as an investigative aid to locations where, ordinarily, we may have lost any corroborative video (that might have been overwritten) by the time we may have gotten there."

It also lessened the likelihood of defence lawyers challenging the video evidence, "because we were able to corroborate so much as to who these people were," Shankaran said.

According to an internal report to the police oversight board, Toronto police conducted 1,516 facial recognition searches between March and December 2018, resulting in potential candidates being selected for 60 per cent of the images being searched. Eighty per cent of those led to the identification of "the criminals responsible for these criminal offences," including four homicides, sex assaults, armed robberies and gang-related crimes, the report said.

Betsy Powell is a Toronto-based reporter covering crime and courts. Follow her on Twitter: @powellbetsy

External Content
Source RSS or Atom Feed
Feed Location https://www.thespec.com/rss/article?category=news
Feed Title
Feed Link https://www.thespec.com/
Reply 0 comments