Computer Vision based 3D model floor construction for Smart Parking System

Abstract

A Smart Parking system has a lot of components such as an automated parking infrastructure, sensors, and a navigation system. For the implementation of the navigation system in smart parking, a 3D floor map is required. A 3D view of maps is always better than traditional maps, but making a 3D model comes at a cost and requires specialized tools. Infrastructures such as hospitals, offices usually have the little luxury when it comes to maintaining their parking spaces, and the proposed system is providing a simple yet effective solution for this problem in this paper. Till now, images are two-dimensional, and tools like Lidar or Kinect are used to get the depth element right. However, to make the floor construction handy, portable, and lightweight a smartphone image-based approach is proposed here to make a 3D model of indoor parking lots. The pillars and the separation walls between parking spaces are easy to identify using deep learning models. A convolution neural network-based architecture was used for object detection. The main problem that remains is to calculate the depth of the objects in the image. Here in this paper, a successful approach is proposed to overcome the problem of finding depth in images.