Comprehensive Study on Object Detection for Autonomous Vehicles to Avoid Collision Risk

Abstract

The proliferation of autonomous vehicle necessitates robust perception systems capable of functioning within inherently dynamic traffic environments. This dissertation systematically evaluates state-of-the-art object detection methodologies for autonomous vehicles through comprehensive taxonomic analysis. The prime objectives encompass: establishing a comparative framework for evaluating heterogeneous detection approaches; identifying domain-specific optimization strategies; and elucidating research trajectories within this rapidly evolving domain. The investigation categorizes detection paradigms across multiple dimensions: spatial representation (2D/3D), computational architecture (single/multi-stage algorithms), and sensory modalities (camera/LiDAR/fusion approaches). Methodologically, this work employs systematic review protocols incorporating bibliometric analysis of high-impact publications from Scopus, Elsevier, Springer, and Google Scholar (2018-2025). Performance evaluation integrates multiple benchmarks utilizing standardized metrics (mAP, IoU, latency) to facilitate cross-architectural comparison. Implementation will adapt high-performing architectures through transfer learning with emphasis on edge-optimization techniques including quantization-aware training and neural architecture search. Validation will employ both simulation environments and instrumented vehicle testing. This investigation addresses critical gaps in perception system robustness while providing actionable insights for autonomous driving technology advancement under diverse operational scenarios, ultimately enhancing traffic safety through more reliable obstacle detection