A Study on Vehicle Detection through Aerial Images: Various Challenges, Issues and Applications

Nowadays vehicle detection and counting at the border of countries, as well as states/cities, has become popular through aerial images because of security concerns. It will play a vital role to reduce the various crimes i.e. (children kidnapping, drug/alcohol smuggling, traffic misconduct, weapons smuggling, sexual misconduct and mission of country-related crime, etc.) at the border of the cities as well as countries. Vehicle detection and counting have various other applications like traffic management, parking allotment, tracking the rescue vehicle in hill areas, digital watermarking, vehicle tracking at the toll plaza and urban planning, etc. However, vehicle detection and counting task are very challenging and difficult because of the complex background, the small size of the vehicle, other similar visual appearance objects, distance, etc. Till now, traditional methodology introduced several robust algorithms which has limitations while extracting the features from aerial images. Recently, deep learning-based algorithms introduced and the outcomes of these algorithms are robust for such kind of applications in the area of computer vision. But accuracy of these algorithms is not optimized in aerial images because the deep learning algorithm required a huge amount of data to train the machine and the size of the object in aerial images is also too small. All these factors affecting the efficiency of the real-time device. This paper provides a brief description of traditional algorithms as well as machine learning and deep learning concepts to identifying the object through aerial images. The study has shown the comprehensive analysis of benchmark datasets and their parameters and corresponding challenges used by researchers and scientists in the area of object detection/tracking through aerial images.