Smart City Crime Detection and Alert System Based on Edge Computing

The recent urban street vending boom has come with consequences in the form of urban disorganization, congestion and health risk. The conventional approach of monitoring street vending, i.e., manual patrols and spasmodic crackdowns by security agencies, has proven to be ineffective owing to high costs, lack of adequate staff, inconsistent enforcement and inability to span a wide geographical area. The aim of this study is to develop a "Smart City Crime Detection and Alert System using Edge Computing and Convolutional Neural Networks (CNN)" to identify, detect, and report illegal street vendors in real-time. The system combines surveillance cameras and image recognition, GPS tracking, and automatic levelling through mobile networks. After being tested on a simulated dataset, the CNN model achieved 92% detection accuracy, response time reduced by 70%, and bandwidth usage reduced by 60% compared to cloud-based methods. Environmental noise and hardware limitations were mitigated by applying model pruning, quantization, and knowledge distillation. The proposed 3-tier hybrid architecture (edge, communication, and cloud layers) facilitates scalability, cost-competitiveness, and privacy control. Future studies will include expanding data sets and adding advanced sensors for broader application.