Real-Time Object Detection Performance of YOLOv8 Models for Self-Driving Cars in a Mixed Traffic Environment

Self-driving cars have gained significant attention in recent years. Real-time object detection is a critical component of their perception system. One of the main challenges in developing safe and efficient self-driving cars lies in accurately and in real-time detecting objects in diverse and complex traffic environments. This paper presents the performance evaluation of real-time object detection of YOLOv8 models, a state-of-the-art deep learning framework for self-driving cars in mixed traffic environments. The objective is to assess the extent of YOLOv8's object detection capabilities can be used for self-driving car within complex real-world traffic scenarios. The experimental results show that the accuracy of YOLOv8 during normal daylight scenarios ranges from $0.60 \sim 0.80$. In contrast, the accuracy values obtained in night scenarios fall between $0.15 \sim 0.25$. Similarly, the F-Measure of YOLOv8 models under daylight conditions ranges from $0.75 \sim 0.87$. Conversely, the F-Measure in night conditions falls between $0.27 \sim 0.46$. Based on our thorough evaluation results, YOLOv8 has demonstrated its robustness in detecting objects. However, the algorithm requires improvement to effectively handle the challenges of self-driving cars' object detection systems in mixed traffic, such as diverse object classifications, small-scale objects, fast-moving objects, blur, glare, and low-light illumination, particularly in night-time scenarios.