A Context-Aware Object Detection Method for Self-Driving Vehicles

EasyChair Preprint no. 9522

Abstract

Some issues with today's mobility are road accidents, fuel inefficiency, traffic congestion, etc., which directly or indirectly affect our economies and lives. AVs, also known as self-driving vehicles, can address these issues effectively. AVs take leverage from sophisticated sensor technologies. They classified perception systems for AVs into two classes: driving environment perception and positioning perception systems. In this paper, we addressed the driving environment perception problem. AVs need to drive safely through the roads. To avoid collisions, they need to identify various objects around them accurately. Therefore, we need a method that detects these objects with higher accuracy. Recent crashes of Tesla, Toyota, and Google self-driving cars indicate that a lot is required in order to make object detection methods for Avs. Hence there is massive scope for improvement. In the proposed work, we addressed object detection for AVs. The proposed object detection is a multiclass image segmentation problem. We used deep learning-based methods. The proposed method can be divided into (1) identifying context and (2) using context-based models for object detection. For performance evaluation, we used accuracy percentage, sensitivity, and specificity. The proposed method showed promising results at par with other schemes. The average prediction accuracy for ten classical deep-learning image frames is 89.021%. The average prediction accuracy for ten image frames of the proposed context-based deep learning model is 95.344%. We can see that the proposed context-based deep learning model produced 6.323% better accuracy than the base scheme.

Keyphrases: autonomous driving, context-awareness, deep learning, object detection

@Booklet{EasyChair:9522,
  author = {Basha Alsahli and Olaa Motwalli},
  title = {A Context-Aware Object Detection Method for Self-Driving Vehicles},
  howpublished = {EasyChair Preprint no. 9522},

  year = {EasyChair, 2023}}