YOLO-V7 and YOLO-V8 Benchmark for Firearm Detection and Deep Learning Model Retraining

Ecuador is currently facing a high level of uncertainty in terms of security, due to a worrying increase in crime rates. A study conducted in 2023 has revealed concerning statistics, with 6,300 homicides and 6,316 armed robberies recorded, making the country one of the most violence-ridden in Latin America. As a result, commercial establishments are becoming increasingly susceptible to security breaches, causing many to install video surveillance systems as a protective measure. However, these conventional surveillance systems are only retrospective tools, providing evidence after an incident has already occurred. To address this urgent security need, a proactive system has been developed to provide real-time notifications upon firearm detection. The system employs a Benchmarker approach, comparing the performance of Yolo v7 and Yolo v8, and implements an IoT architecture for AI model retraining, utilizing Amazon and Google Colab services. Time is a crucial factor in the effectiveness of the detection system, so Yolo v8 was chosen due to its 21% improvement in processing time, but at the cost of a 20% increase in computational demands (RAM and GPU). Additionally, the system places a strong emphasis on using cameras that are compatible with the processing card, rather than relying on CLOUD-based streaming services, which has resulted in a notable 71% reduction in latency, enhancing the system’s responsiveness.