Learning-based Auto Focus

Speed up the focus speed of Speed Dome cameras using machine learning.

Learning-based Auto Focus technology is used in selected ACTi Speed Dome cameras to increase focusing speed after panning, tilting or zooming.

Conventional focusing techniques rely on detecting peaks in contrast and sharpness of objects' edges in the scene while continuously adjusting the focus. The image is becoming sharper beyond the point at which the focus is optimal. Once this point has been exceeded, the lens returns slightly to the position with the sharpest image. To the viewer, the image appears sharp, then blurry and then sharp again. Depending on the lighting conditions and other factors, this process may take up to several seconds.

Once in focus, ACTi Speed Dome models equipped with Learning-based Auto Focus technology use reverse calculations to determine the distance of the object in front of the camera at each specific position (defined by pan and tilt coordinates) and store it into their internal memory as a reference point. Next time the camera moves to this position, it uses the previously saved information about target's distance to immediately focus on it. The action takes a split second, significantly less than if the camera had to find focus without any reference.

In a typical scene, the majority of objects are static, like furniture, buildings, or trees. Their distance from the camera is unlikely to change over time. However, some objects are mobile, for instance, people or vehicles. It is possible that an object's distance that had been found previously at the position is no longer valid. In that case, the camera detects suboptimal image sharpness and automatically starts refocusing. Since the camera has a reference distance, it attempts to find the optimal focus around the previously measured distance. It is able to refocus nearly instantly because it doesn't need to search focus in a range between few centimeters and infinity. It then updates the focus information for this position in its memory.

The learning is completely autonomous and continuous, resulting in the camera's focus becoming faster over time simply by using it. No action needs to be taken by the camera operator, although defining the focus manually for any position is also possible. The focus created manually then takes precedence over the automatically learned one.

Related Products

Resources