Autonomous Weapon Detection System in Real-Time Monitoring Using Convolutional Neural Network

Background-This study delves into convolutional neural networks (CNNs) and the TensorFlow Object Detection API. The research utilizes these advancements to retrain a pre-existing model for real-time surveillance weapon detection. Methods- The objective is to examine the impact of training CNNs with an additional class, "non-weapon," alongside the original classes, "gun" and "knife." The goal is to determine if the model's accuracy varies when identifying non-weapon objects. Results- The research findings demonstrated that utilizing an Inception feature extractor enabled the faster R-CNN model to achieve a commendable mean average precision of 95.50% in identifying gun, non-weapon classes and knife. Concluding Remarks- Soldiers are uncomfortable since the temperature is changing daily. Soldiers are affected by severe health issues due to the sudden changes in climate conditions. So, in this work, the body temperature of soldiers and environmental weather states are sensed by using sensors, and based on these sensed values, the body temperature is regulated automatically. The measured parameters are displayed on the mobile screen through the Wi-Fi module NodeMcu, which is interfaced with a temperature sensor and a Peltier module and is the major element in this research.