Deep Learning Based Traffic Surveillance System For Missing and Suspicious Car Detection
Vehicle theft is arguably one of the fastest-growing types of crime in India. In some of the urban areas, vehicle theft cases are believed to be around 100 each day. Identification of stolen vehicles in such precarious scenarios is not possible using traditional methods like manual checking and radio frequency identification(RFID) based technologies. This paper presents a deep learning based automatic traffic surveillance system for the detection of stolen/suspicious cars from the closed circuit television(CCTV) camera footage. It mainly comprises of four parts: Select-Detector, Image Quality Enhancer, Image Transformer, and Smart Recognizer. The Select-Detector is used for extracting the frames containing vehicles and to detect the license plates much efficiently with minimum time complexity. The quality of the license plates is then enhanced using Image Quality Enhancer which uses pix2pix generative adversarial network(GAN) for enhancing the license plates that are affected by temporal changes like low light, shadow, etc. Image Transformer is used to tackle the problem of inefficient recognition of license plates which are not horizontal(which are at an angle) by transforming the license plate to different levels of rotation and cropping. Smart Recognizer recognizes the license plate number using Tesseract optical character recognition(OCR) and corrects the wrongly recognized characters using Error-Detector. The effectiveness of the proposed approach is tested on the government's CCTV camera footage, which resulted in identifying the stolen/suspicious cars with an accuracy of 87%.