Multilayer vehicle classification integrated with single frame optimized object detection framework using CNN based deep learning architecture

Here we have rendered a functional and architectural model of a system that assists the driver of a vehicle to detect, identify and track objects while driving. The objects detected include vehicle type as well as common on-road objects such as pedestrians. Layer structure for the system involves the design of a state-of-the-art deep learning classifier using a novel database for obtaining higher classification accuracy and another layer consisting of a single-frame object detection method to make the system more robust while limiting the processing time involved. Sub-systems integrated to facilitate the driver with relevant real-time information about his driving umwelt include vehicle identifier, number plate recognition system and creation of database consisting of collected information along with time-stamp. Performance degradation under various ambient conditions and variable environments with various synthetic noises being introduced in the video frames have been studied. Trade-off between speed and accuracy of a state-of-the-art real-time detection system implemented on various processing platforms is studied. Layers of deep learning classifier were trained using an optimized dataset consisting of static and dynamic images of vehicles to yield suitable prediction accuracy and this was combined with a system pre-trained on COCO dataset for YOLO. This helped complete the Intelligent Driver Assistant System. This paper also includes the implementation of real-time object detection on a single board computer. This concept can be tapped to create compact and portable driver assistant systems.