This chapter details initially acquiring an open-source UAV dataset and creating a Google Earth dataset of vehicles, and creating the metadata for these images. Then training a deep learning object detection model, YOLOv4, to generate the best training weight files, having a very high mean average precision (mAP). It is the measure of how precisely the model is detecting the objects specified in the metadata of the validation dataset. The higher this value, the more accurate the model is, with the specific data it has been trained upon. Then deploying the model on various satellite data of certain areas like parking lots, toll booths, and roads, over a certain period of time, to count the number of vehicles in RGB images and from those images calculate factors like maximum capacity of parking lots, average vehicle density of roads, congestion rate in toll booths, length of congestion in toll booths, etc. The model trained on the UAV Dataset at various conditions of weather, daytime, and different resolutions are tested over other UAV Datasets and the trained weights are uploaded to GitHub for future use.
TopIntroduction
Geospatial data is very important because it helps us better understand the planet that we live in. It is the data collected from various satellites/UAVs that orbit/hover this planet. The reflected/emitted electro-magnetic rays by different surfaces help us better understand and classify various characteristics of the surface features. This data is collected by the sensors on various satellites/UAVs, which are designed to operate in different parts of the electro-magnetic spectrum. The geospatial data in the form of images consists of matrices of pixels with reflectance values that are more easily understood and utilized with the help of various mathematical functions and filters applied to the data. However, no matter what we do with this data, the final inference has to be derived with the help of our cognitive abilities.
The development of computer vision is a field of artificial intelligence (AI) that allows computers to derive useful information from digital images and videos. From an engineering perspective, it tries to understand and imitate certain functionalities of the human ocular system, and automates the task based on inferences made (Douillard, 2021). Machine learning (ML) is a subset of AI which uses computers to learn from data with the help of algorithms, and creates model based on patterns of functionality (Wolfewicz, 2021). It is a cross of computer science and statistics, where mathematical functions are used to predict certain required parameters from a set of data. Computer vision algorithms based on ML are Linear, Polynomial, and Logistic regression, Support vector machines, Decision trees, Random forest, etc. Deep learning (DL) is a further subset of ML which uses computers to learn from data with the help of very complex structures of algorithms modeled on the central nervous system of human beings (Wolfewicz, 2021). The DL algorithms are very sophisticated and mathematically complex that analyze data with a logic structure analogous to how humans logically deduce the information. Computer vision algorithms based on DL are R-CNN, Fast-RCNN, Faster-RCNN, R-FCN, RetinaNet, SSD, YOLO, etc.
Any data which contains information of Earth surface (Garg, 2022), like a district, a small parking lot or a road or a toll plaza, etc., fall under geospatial data. The information acquired from this data can be used for future technological developments, or better use of Earth resources, or for the advancement of the human civilization. Vehicles that we use now are the culmination of technology. Due to the growing concern for climate change and the drastic shift of the balance between the biomass of the Earth, endangering life because of the harmful greenhouse gasses emitted by vehicles, the car manufacturing companies has pushed the global markets to build and sell more electric vehicles (EVs). With more and more people shifting towards EVs (Malmgren, 2016), the requirements of data related to everyday vehicle movement, parking, and the functioning of traffic need to be studied and analysed. The geospatial data along with DL can be used to predict available vacant slots in parking lots, or duration of clearance on a road based on the average vehicle density of the road and mean congestion rate, or duration of clearance of traffic from a toll plaza, which will influence the decisions of human-beings concerned with parking their vehicles or reaching somewhere on time.
In this chapter, we discuss a specific DL algorithm which performs a specific task of detecting objects (vehicles) from an image and classifying them, along with certain functions that count the number of specific object (vehicle) or all objects (vehicles) and extract characters from the license plate of vehicles. The YOLOv4 algorithm, which is a DL one-stage object detection algorithm, is used to detect objects from digital satellite images (Landsat-8 and UAV) according to its training to detect from data source.