A Real-Time Traffic Surveillance and Security System using Transfer Learning and Edge Computing

Two Wheeler's are one of the most commonly used vehicles worldwide, and the helmet is the most critical gear/safety equipment. There is no proper system that validates whether all the users are equipped with a helmet while they are riding. The existing traffic surveillance and security systems are designed only to identify the vehicles that over speed and jump over the red lights. These systems use interceptors at highways to capture the violators and consume a lot of manpower and vehicles for patrolling. An efficient traffic surveillance system to monitor whether the motorcyclist and the pillion rider is wearing a helmet is missing. This paper proposes an efficient and reliable traffic surveillance system using edge computing and transfer learning that continuously monitors the two-Wheeler's and sends alerts and information on the non-helmet riders to the nearest police vehicles. The edge computing is being applied at the input/entry-level (CCTV camera) and helps in reducing the delay of the entire system. The classification models are created by applying Transfer Learning on ResNet50, VGG16 architectures, and applied to real-time video footage from the CCTV camera to detect non-helmet riders. The proposed system gives an accuracy of 97.50%, reduces the manpower, and enables the police interceptors to work efficiently.