Visual analytics using artificial intelligence (object detection in car cabin)

Abstract

The development of computer vision technology has promoted the automobile industry to rapidly turn to autonomous driving technology. It is becoming more and more important for vehicles to accurately detect and interpret the objects in the cabin. Object detection is the core of this capability. Under these circumstances, the object detection based on deep learning has made great progress. Currently, it is mainly segmented into two-stage methods and one-stage methods. YOLO, with its simple algorithmic structure, high detection accuracy and fast speed, has become a representative of one-stage methods and received widespread attention. Since YOLO was first proposed in 2015, researchers in the field have made many iterations to update YOLO, making its performance more and more powerful. The purpose of this project is to detect objects in the car cabin through the YOLO algorithms. Considering the illumination, shadows and other factors, the datasets used to simulate the real interior situation include IR images and RGB images, and they are tested with YOLOv8 respectively. Then all RGB and IR images are mixed to build a new dataset for training. Based on the results, compare the mean average precision and training speed of the model across the three datasets.