Real-Time Parking Lot Occupancy Detection System with VGG16 Deep Neural Network using Decentralized Processing for Public, Private Parking Facilities

Efforts have been taken to develop an efficient image-based parking occupancy detection system employing the camera systems (CCTV) pre installed for security purposes. The present detection system has experienced the problem of identifying the subject with high accuracy and applicability in real time environments. A transfer learning methodology where CNN's feature extraction was used with binary classifier has proved to provide optimum accuracy. This paper proposes a decentralized system to detect the occupancy status of a parking slot from the CCTV video footage using VGG16 CNN architecture for feature extraction followed by binary classification to detect vacant or occupied spaces. To deliver a complete parking solution, the model has been tested on Live CCTV footages collected from existing parking slots and have provided a complete integrated system that allows real time availability check and booking of parking slots. This paper provides a detailed explanation of the execution of the above system. The performance efficiency and detection are improved by testing the designed model on two datasets Pklot and CNRPark, CNRParkEXT comprising daylight images with multiple weather conditions. The proposed technique is compared with the current state of the art implementations [1],[2]. The dataset has been developed to test the model under low intensity light conditions. With significant accuracy and high-performance throughput observed, the proposed system can be effectively implemented at the commercial level.