Computer Vision

ObjectVideo was one of the earliest commercial innovators in the field of Computer Vision. ObjectVideo's development efforts today address the far-reaching implications of ubiquitous video souces, from hand-held devices to ever-expanding IP video deployments in public spaces. Any effective analytical framework must now accommodate greater variations in video source placement, simultaneously track multiple targets, and effectively manage large volumes of data.

ObjectVideo in Action


Airborne Tracking
(Multiple Targets)

Airborne tracking of multiple objects is challenging due to several factors, including extreme camera jitter, motion parallax, and the small size of objects viewed from a high altitude. ObjectVideo airborne algorithms compensate for these challenges,  effectively tracking multiple objects simultaneously.

Airborne Tracking
(Target Paths)

ObjectVideo airborne algorithms track both people and vehicles and transmit target position and trajectory data in real-time.


Adding Structure to Video

Video data is hard to search. ObjectVideo video-to-text algorithms render video more easily searchable.
In this example, as objects move through the camera view the video analysis engine returns text describing object type, position, trajectory, and type of activity.

 Vehicle Model Identification

License plate recognition is limited in that it only works under certain constrained environments. ObjectVideo vehicle matching algorithms can augment vehicular identification systems by using other classification and identification criteria, such as vehicle shape and size.


Target Tracking Across Multiple Sensors

Video analytics solutions must be able to correctly classify and track targets within an ever-widening range of video formats, lens types, and hardware features and then seamlessly ingest and meaningfully analyze volumes of data. The following example shows target trajectory tracking across multiple sensors with differing fields of view.


For airborne vehicles monitoring an area, the geospatial data of objects in that area is frequently imprecise. OV has developed a variety of approaches to mitigate this imprecision, allowing real-time overlay of GIS data (LIDAR, road networks, compass, etc.).


Video to Video Search

This demo illustrates how video clips may be submitted for contact analysis and matching. Each query is annotated according to object and event metadata (e.g., people dismounting from a vehicle, vehicles passing, people passing inanimate objects between one another, etc.). Upon query submission, a set of candidate videos is reviewed, a list of possible matches is returned, and that list is sorted according to likelihood of a positive match.