Small Object Detection in Autonomous Cars Using a Deep Learning

Abstract

In computer vision, object detection plays a vital role in ensuring the safety of a self-driving car. The most notable example of this continuing exploration and enhancement in the field is the Google Self-Driving Car Project, currently known as Waymo. Although many prior studies have utilised several object detection models to improve the efficiency of autonomous cars, each comes with its own set of challenges. The major challenge in the development of self-driving cars is latency. The latency refers to the delay between the processing of input data captured from the camera and the decision taken by the machine learning algorithm to move and direct the car on a safe road. To address these issues, we propose a MobileNet SSD framework by hyperlinking MobileNet with SSD, making it sufficient for real-time applications. The model utilises two types of sparable convolutions, namely spatial separable convolutions and stepwise sparable convolutions. The result demonstrates the efficiency of our proposed MobileNet SSD model in reducing computational costs and decreasing latency.