Vision Based Real-Time Active Protection System Using Deep Convolutional Neural Network

Abstract

The health of children and women are highly affected due to conflicts or war. The effects of war create terrible emotional consequences and physical consequences. The well-being and development of nation are also ensured by an intelligent defense system. Threats faced by tanks and other armored vehicles on the battlefield are getting more complicated. The proposed vision based active protection system installed in the tank is capable of recognizing the hostile targets precisely and destroy targets in the air before entering into the territory. This real-time Active Protection System can save the life of the civilians during warfare. The proposed model integrates a vision-based image processing technique with ultrasonic sensor for the real-time active protection system. The model utilizes lightweight deep CNN model (YOLOv5s architecture) on a Raspberry-Pi1 processor to recognize the hostile targets. Then, the predicted data is transferred from Raspberry-Pi1 processor to the cloud. Raspberry-Pi2 processor receives the information from the cloud and controls the missile operation of the tank in real-time. The Raspberry Pi processor is a low-power computing device, and YOLOv5s is familiar for its light weight and timely recognition. The proposed YOLOv5s model obtained an Average Precision of 93.10%, Average Recall of 89.50%, and F1-score of 91.26%. The Prediction time of the model is 4.1ms on Google Colab and 405ms on Raspberry-Pi processor.