Unlocking the Potential of Raw Images for Object Detection with YOLOv8 and BOT-SORT Techniques

The basic use case of raw image for object detection and tracking in autonomous driving with the help of YOLOv8,also it uses the BOT-SORT algorithm be discussed in this paper. Converting raw sensor data to processed images using Image Signal Processing (ISP) is a computationally intensive and expensive task that traditionally has been in the image processing pipelines of many applications. In this paper, we examine these costs and the processing pipeline itself and propose a solution that eliminates them by consuming raw images directly. We perform our experiments on the RobotCar dataset, which suggest that raw images achieve competitive precision, recall and F1 scores to their preprocessed counterparts but in a fraction of the time. This work validates that raw images are a feasible and advantageous way to perform real-time object detection and tracking, contributing as possible alternative for coachbuilders. This work verifies by experiments the efficiency of raw images applying YOLOv8 and BOT-SORT in-depth, which significantly reduces cost and time over other related researches as a primary contribution. Our work on the findings studied indicate that raw images lend a direction toward efficacy and performance for improving autonomous driving system.