Crying Wolf
False alarms arise for a number of reasons. For example, typical sources of false positives for people counting are shadows from people walking outside the field of view of the camera, and these can be minimized by proper positioning of lighting, Ashani said. “False negatives might occur when an adult and child are passing through and walking very close together. The software may mistake them as one person. There is no effective way to get around this, but statistically speaking, this situation is rare enough.”
Usually, high false alarms can stem from wrong applications of the technology, poor products, bad installations and poor understanding, Lim said. “No verticals should ever accept frequent false alarms; thus, the first three factors must be thoroughly tested in proof of concepts or pilot projects.”
Proper definition of false alarms is also a significant challenge, Lim continued. “It is normal that environmental factors such as trees and weather can cause some false alarms, but these should have minimal effects for a properly calibrated system. However, if a suspect's action is detected as a potential intrusion, though it did not happen, it should not be determined as a false alarm.” When a system constantly generates alarms when there is no good reason to do so, “operators cease to treat them as critical situations; they begin to dismiss or ignore alarms without proper verification,” Geutebrück said. “Then, the human operator is back to square one. The system ends up providing nothing but a false sense of security.”
The Eyes
While the analytics are the brains, the cameras are the eyes. As with human vision, the healthier the eyes, the better the vision; the advent of HD and megapixel cameras greatly increases the practicality of VCA. “We are always delighted when our customers choose manufacturers whose products provide consistently good images required for video analytics,” Li said. “We sometimes cringe when the customer insists on using certain brands which we know will cause problems because they were not designed with VCA in mind.” When customers insist on using older analog cameras, integrators and installers face difficulties in achieving the full suite of VCA features, Lim agreed. “Megapixel cameras definitely provide more possibilities to apply the VCA features.”
Camera Placement
Appropriate camera placement is also important. For example, people counting is impractical when the camera is not mounted correctly, Ashani said. “A low hanging camera mounted at an angle will make the count very inaccurate as people passing through the field of view will be blocking each other.”
A top-down view at the right height can provide maximum accuracy for people counting, Li added. Another example could be to detect a person moving toward the camera, Ashani said. “A camera that is almost parallel to the ground will detect little directional movement.”
Staying in the Loop
An often ignored element is the work flow. Even if VCA were 100-percent accurate — which is currently impossible — work flows must also be in place to support an alarm, Chandler said.
“For instance, we have the ability to use thermal cameras to detect someone fishing or poaching on a reef in the middle of the night, and our VMS can notify customers via alarm,” Chandler continued. “But who verifies the alarm? Who is dispatched? Often times, there are no blueprints for how to deal with the ever-growing amount of information VCA can provide, so we help policy makers understand options and change frameworks to support these new types of metrics and alarms. We feel the landscape is changing, and that VCA adoption will continue to grow as customer response frameworks better support what VCA can provide.” An alarm may be a digital signal, but the coordinated response should not be based on the simple yes/ no/true/false questions that VCA answers, Chandler added. “The goal is to put supporting information back into the hands of decision support managers who must ultimately make the call.” VCA cannot save lives in our world — humans do!