Image Recognition Using Yolov5 for Automotive Industry

Abstract

The automotive industry greatly benefits from the implementation of a robust object recognition system, which enhances the efficiency of the inspection process. This paper introduces a GUI system designed to address issues related to operators mistakenly attaching incorrect barcodes to car seats, leading to misplacement in the production line. The study aims to achieve two objectives: firstly, to implement a deep learning framework based on YOLOv5 for car seat pattern recognition, and secondly, to develop a GUI interface that utilizes camera inputs to monitor car seat types and evaluate the performance of the YOLOv5 detection model. To achieve this, the project utilizes the YOLOv5 algorithm for car seat detection, employing a custom dataset comprising three types of car seats: Type A, Type B, and Type C. The dataset is labelled accordingly using the Roboflow website, and Google Colab is utilized for training the custom dataset, which is done over 90 epochs.  Subsequently, the GUI system is developed using Tkinter, which enables car seat detection from both static and real-time images. The results obtained from the GUI system showcase the successful detection of car seats based on static and real-time inputs. The average confidence scores for Type A, Type B, and Type C, using the static images method, are found to be 0.53, 0.64, and 0.61, respectively. On the other hand, the average confidence scores using real-time images are slightly lower, with Type A at 0.30, Type B at 0.53, and Type C at 0.38.  Upon comparing the results from static and real-time images, it becomes evident that the static image approach yields more accurate detections compared to the real-time image method.