Object Detection for Autonomous Vehicles in Adverse Weather and Varying Lighting Conditions Using a Hybrid YOLO Approach

Object Detection is a major task in Computer Vision with applications ranging from Surveillance to Autonomous Vehicles. In the detection process, manually annotating images would be more labor-intensive and time-consuming particularly if we have a large dataset. To overcome this, the YOLOv9 model is employed as an annotation technique to automate image labeling that accelerates the labeling process. The YOLOv8 model is then used for model training and inference to detect objects. YOLOv9 could take 9 minutes and 23 seconds to generate class-labels for around 4.8K images. YOLOv8 efficiently detected objects across five classes - Cow, person, car, truck and dog. This illustrated how semi-automated annotation can significantly reduce labeling time and effort on custom datasets. It was observed that the YOLOv8 model achieved good performance with a mAP.50 of 84.5% and a mAP.50-95 of 70.1%. This demonstrates that the hybrid YOLO approach is well-suited for real-time object detection.