SA5 | Devpost

training
image_visualization_1
testing_1
testing_2

Problem Statement:

Artificial Intelligence Monitoring systems, are required to convert the legacy surveillance camera inputs, into smart inputs using AI.

Existing Solution:

The legacy systems currently used to achieve battlefield transparency, need to be continuously manned and monitored. This poses a problem of inefficient management for manpower, for resource manipulation. Thus, an innovative adaption is required to help manage human resources, without compromising on the level of security deployed.

Proposed Solution:

We have proposed a solution for this, using COTS Surveillance cameras, and AI monitored surveillance techniques. We have integrated highly accurate Infrared thermal imaging cameras and sensors, in our Commercially off-the-shelf surveillance device, that will provide continuous real-time surveillance data inputs. These data inputs will then be fed to our Neural network which utilizes MASK R CNN algorithms to thoroughly identify any and all threats in the video feed. This provides a cost-efficient and highly accurate means of achieving complete battlefield transparency.

Conclusion:

While Implementing this technique using a small dataset, we achieved results where even partial hands were detected. On further training using custom COCO datasets, we have achieved better results and higher accuracies.

Future Scope:

In the long run, this may be deployed using advanced nanosensors and miniature drones to provide in-depth surveillance over any suspicious region. The MASK R CNN AI detector can be improvised with other advanced algorithms to provide better defense against hostile attacks.