Thrown Into the Deep End

Oct. 27, 2008
Intelligent video system brings poolside safety into the 21st century

In 2003, the National Center for Injury Prevention and Control in Atlanta recorded 3,306 unintentional fatal drownings in the United States . Many of the victims were children who drowned in both public and private pools. Regardless of the locale, a common thread ties the tragedies together: pool drownings are preventable. But even with trained lifeguards and aquatic safety awareness, accidents happen.

Contrary to popular belief, drowning is a silent killer. A swimmer in trouble is unlikely to panic and splash about; immobilized by shock, the individual cannot move or breathe, let alone cry for help. The response-time is crucial – a few seconds can make all the difference between recovery and permanent brain damage, between life and death. If lifeguards or other emergency personnel intervene and begin resuscitation within 30 seconds, the victim's chances of recovery are very good, but lifeguards are human -- and that means that they cannot see every swimmer 100 percent of the time. Add heat, noise and light playing with the waves, sometimes trouble remains hidden.

We have smoke detectors in homes. We have surveillance cameras in shopping malls. We have air traffic control systems to keep airplanes from flying into each other. Where are the tools for lifeguards?

Seven years ago a French company was founded on the idea that machine vision technology could help prevent pool drownings; that the right mix of cameras, computers and software could ensure swimmers' safety. Today, Vision IQ/Poseidon develops a computer-aided drowning detection system that helps lifeguards save lives.

The Poseidon System

Although the configuration varies according to the requirements of the pool, each Poseidon system uses the same components: underwater cameras, overhead cameras, PCs, a Matrox Morphis Dual or Quad frame grabber, an LED display panel and a waterproof-touch screen supervision monitor.

Depending on the size of the pool, the Poseidon system may have between 8 and 32 cameras networked to one or more computers. In a typical indoor configuration, overhead cameras will be installed on the ceiling, and if the pool is deeper than nine feet, underwater cameras will be installed on the pool's wall. A Matrox Morphis Dual or Quad frame grabber digitizes the cameras' analog output so the image data can be analyzed. First, the Poseidon software notes if a swimmer's behavior is unusual, and then tracks the individual for several seconds. If the software detects a motionless presence on the bottom of the pool for 10 seconds, an alarm notifies the lifeguards and the LED panel displays the victim's coordinates in the pool. This panel is also equipped with red lights and a siren that are activated when the system detects an incident.

The frame grabbers acquire image data continuously while the software analyzes each image and compares it to other images captured by the remaining cameras. “Detecting someone in trouble is entirely vision-based, that is, it is the result of passing the algorithms on the images that tell us when someone is having difficulty in the water,” explains Francois Marmion, Vision IQ general manager. In real drowning cases, people sink very quickly. Furthermore, the system's algorithms can detect the difference between someone who is bobbing above the water and falling under the water. “We compare the images acquired from overhead and underwater cameras, and using all these ways, we can tell that someone is in trouble,” Marmion says.

In addition to developing specialized software, Poseidon has also designed its own custom hardware. Poseidon's team designed a waterproof monitor to display the user interface. Often the pool managers opt for a poolside monitor, so it must function in an environment with heat, humidity and chlorine. Lifeguards must be able to touch it, even with wet hands.

The underwater camera used in the Poseidon system is also custom-designed. It requires a wide angle and waterproof housing strong enough to withstand the pressure of deep water.

Overcoming the challenge

Six years of development went into the software. “We have algorithms that determine the volume and the texture of an object in the water,” Marmion says, “and we've established the criteria for when to interpret that object as a person. We also have algorithms to compare consecutive images that can check the position of that object.”

The most difficult task, however, was managing the effect of light on the water's surface. Marmion says that architecture has made an impact on both lifeguarding and Poseidon's development. Modern pool design features light, with larger windows to let in more natural light. This offers a nice environment for the swimmers, but from a safety point of view, the light interferes with the lifeguard's view. Since the cameras are extra eyes for the lifeguard, the software must be able to ‘see' the difference between real objects and light effects, just like the lifeguard.

Extensive testing was vital to overcoming the software challenges. “We had a steep learning curve,” Marmion says. Poseidon's team set up two pilot sites in France , where the system operated constantly and provided volumes of data for analysis. Mannequins simulated actual drownings. “Thanks to the pilot sites, we had lots of results and we tested the software, little by little,” Marmion says.

Another challenge for Poseidon was determining the threshold between detecting someone in the water and a false alarm. The alarms are set to go off after 10 seconds. Using a shorter time frame will send too many alerts, and that might not help the lifeguards do their job. “They want the alerts when it's really an important event,” Marmion says. “Finding the right threshold was part of the technical challenge. Ten seconds is the right number to give the victim the best chance of recovery.”

Currently, Poseidon has 50 installations in France, 50 in the rest of Europe, and three in Japan . An additional 50 systems have been installed in the United States , mainly in YMCAs.

For more information, visit .

Since 1998, Sarah Sookman has worked at Matrox Imaging as both a technical and marketing writer. For more information, visit