Detection and Identification of Vehicles from High-Resolution Aerial Images Using Deep Learning Approaches with the Tuned Parameters

Unmanned aerial vehicles have been utilized in various applications for the management of the civil and defense infrastructure. These applications include the surveillance of the crowd, transport, and disaster management and the inspection of the building architecture. This excessive use of the unmanned aerial vehicles also demands the devices and the deployed system that devices are fully autonomous, efficient, and accurate. Classification of vehicle type is the major component of intelligent transportation system. Traditional approaches like (Bayesian prior models, scale-invariant feature transform, speeded-up robust features, and support vector machine) image processing and machine learning are used by many researchers to automate the transportation system. However, they failed to provide the promising results because of limited viewpoints and lack of dataset. With the advancement in technology, deep learning has inspired us all with its autonomous feature extraction and accuracy. In this chapter we present a deep learning-based approach which involves two phases: region proposal network and classification network; both networks use the features from convolution neural network (CNN). Proposed system first detects the vehicle and then classifies its brand type accurately on self-generated dataset with the help of UAVs. Proposed system classifies the vehicles with the average accuracy of 96.2% on six types of vehicles.