AI-Enabled Drones Track “Out of View” Enemy Tanks

VIDEO: Army& Raytheon Build New AI-Empowered EW “Jamming” System

By Kris Osborn, Warrior Maven

(Washington, D.C.) Should an enemy tank all of a sudden move into and then exit the video “field of view” generated by a surveillance drone, is there any way for human operators to discern an idea about where it might have gone? Or be able to accurately predict where, how and when the enemy might return into a drone sensor’s detection envelope? Establishing a continuous track on a target or object of great tactical relevance can prove difficult for even the most advanced drones. Perhaps the enemy tank enters a cave or thickly wooded area to deliberately obscure itself from detection? Perhaps the enemy tank turns off its engine to remove its heat signature and avoid being found by thermal sensors? Perhaps it moves into structures or behind buildings to diminish or remove its radar signature? Doesn’t this complicate, if not even preclude, successful surveillance and targeting?

These are challenges which the military services and its industry partners have continued to address over a period of many years, and various initiatives are now being quickly improved by new sensing, computing and information processing technologies. Analyzing and disseminating full-motion video from drones is growing in complexity and of course importance, as it is something now generating a large amount of attention from the Pentagon.

As part of the military-industry collaborative efforts to improve full-motion video analysis, the company CACI has developed software intended to essentially enable drones to anticipate or “predict” the trajectory, flight path or movement of a target or object of interest. What if a drone were able to autonomously predict where or when the enemy tank might return into view, or establish an uninterrupted targeting track from human commanders to follow? Drawing upon artificial intelligence (AI) and what’s called deep machine learning, CACI’s emerging software seeks to do this.

“CACI’s software and techniques can track an object persistently. If an object leaves and re-enters the field-of-view of the sensor, our neural network model will automatically detect the object and provide the same unique ID to that object,” Brian No, CACI Manager of AI Research & Development, said in a company essay on AI.

This means that instead of having to piece together segmented or otherwise disconnected portions of video, advanced CACI-engineered computer algorithms can analyze information within a broader context of cues, clues and historical indicators to bounce new information off of a compiled database in seconds to establish an anticipated trajectory for an object, in effect generating a continuous “track” even if a target momentarily leaves a sensor’s field of view. The idea, CACI developers explain, is to go beyond mere object identification and in addition establish an identifiable steady track on a target.