Object Detection Systems for Autonomous Vehicles

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Efficient and accurate object detection is a cornerstone of computer vision applications, particularly in autonomous driving systems where real-time performance is paramount. This study revisits the fundamental challenge of balancing detection speed and accuracy, a critical trade-off that has defined the evolution of computer vision for automated systems. The objective of this research is to empirically evaluate this trade-off and demonstrate a strategic approach to its resolution through architectural design and loss function selection. A lightweight, custom deep learning architecture, termed SimpleNet, was developed and trained on the KITTI object detection benchmark dataset and adopting the Object-Oriented systems analysis and design methodology. The implementation prioritized computational efficiency and was executed using Python and TensorFlow. The core of this research involved a controlled experiment comparing SimpleNet's performance under different configurations, specifically varying network depth and employing different loss functions, L1 and L2. The system's performance was quantified using metrics such as mean Average Precision (mAP) and inference time. While SimpleNet achieved a modest mAP of 12.83%, its primary contribution lies in demonstrating how a streamlined, 3-layer architecture with an L2 loss function can deliver competitive inference speeds (approximately 10.9 frames per second), making it a viable candidate for deployment in resource-constrained environments. This finding provides a practical, foundational case study that highlights the importance of architectural optimization in solving the problem of speed in real-time object detection systems. The results provide a crucial baseline for the design of future lightweight models and contribute to the body of knowledge on design trade offs in computer vision.