PORTLAND, Ore. Ever wonder how blurry surveillance video images can be admissible as evidence in court? Software tools like Sarnoff Corp.'s VideoDetective mine the hidden data in such images to reconstruct their details clearly in still shots. But such tools are affordable only by large corporations and government agencies.
Now a service called Sarensix permits private contractors to "farm out" the forensic evidence they gather from surveillance videos. The service was created by Sarnoff (Princeton, N.J.), a government contractor that fabricates custom ICs and ultrasmall video systems and software. Using data fused from video, infrared and other sensors, Sarnoff's security systems guard government installations and assist troops in the field.
VideoDetective reconstructs video into stills that gather information from many frames, thereby creating sharp, telling still images from indistinct video. Smaller customers can use Sarensix to get their surveillance videos processed in VideoDetective by a Sarnoff-trained professional. "Usually, people have to have our system installed at their site and their own personnel trained to use VideoDetective," said Mike Matisa, a member of the technical staff at Sarnoff. "Sarensix allows customers who have an immediate need, or who can't afford our system, to have us enhance their video for them."
VideoDetective works by first deinterlacing the video so that each frame contains only its own image's information (conventional video interlaces images with scenes from successive frames in order to be compatible with older televisions). Next, the video is enhanced to form a panoramic still image. The image includes all of the details captured by the camera as it sweeps back and forth. Sarnoff subsidiary Pyramid Vision pioneered the panoramic capability for its TerraSight software, which stitches together terrain video coming back from military unmanned aircraft.
The detective part of VideoDetective offers two modes for resolution and enhancement: Super resolution reconstructs details such as license plate numbers, while super enhancement resolves such details as facial characteristics to help make visual identifications.
The picture is reconstructed starting with the identification of a key frame, along with as many other frames as the operators think necessary--either a range of frames on either side of the key frame or a list of individual frames known to have important details in them.
After that, the algorithm aligns the selected frames, then integrates their details into a composite still frame, with stunning details. The technology can identify persons under surveillance even when there is not enough detail in the video for "naked eye" identification. It can also resurrect text.
The service accepts video submission in almost any recorded format, from high-resolution digital video to low-quality surveillance or amateur video.