Vehicle Detection through Instance Segmentation using Mask R-CNN for Intelligent Vehicle System

The recent advancement in artificial intelligence approach or deep learning techniques explored the ways to facilitate automation in various sectors. The application of deep learning with computer vision field has resulted in realization of intelligent systems. Vehicle detection plays a key role in Intelligent Vehicle System and Intelligent Transport System as it assists critical components of these systems like road scene classification, detecting obstacle vehicles to find an unhindered pathway, and even preventing accidents. This paper presents an implementation of Mask R-CNN state-of-the-art method using transfer learning technique for vehicle detection via instance wise segmentation which produces bounding box and object mask simultaneously. As the autonomous systems demands precise and flawless identification of the vehicles thus segmentation based approach is adopted. The model performs satisfactorily for occluded and small sized objects as well. This study is accomplished using an online GPU and cloud services provided by Google Colab by using Tensorflow and Keras framework. A mAP of 90.27% and mAR of 92.38% is achieved by using a combination of benchmark datasets.