Roadside Parking Spaces Image Classification Using Deep Learning

Abstract

The shortage and imbalance of parking spaces have become serious problems in recent years on the roadside. Drivers may park incorrectly, which may take more parking space and disturb other drivers. To mitigate these problems, real-time parking monitoring with a detection system, is very indispensable. In this paper, we propose a classification system of images associated with the parking spaces for vehicles on a roadside based on the convolutional neuron networks (CNN). The image is supposed to be captured by a camera or other visual devices installed on the roadside and sent to a central computer for processing. The treatment is followed by a classification based on DL (deep learning) to determine the occupation of the parking spaces. The proposed classifier model uses fewer calculation parameters that involved a faster model by applying the principle of dilation. The proposed model has been evaluated and compared with the previous models using PKlot dataset, a set of well-known images publicly available. The evaluation results show the reliability of our proposal comparing with the previous works.