Object Detection on Road Using Deep Learning Approach

Abstract

For a developing nation like Bangladesh, traffic jams are a major problem. Systems for maintaining traffic manually are expensive and time-consuming. Making a system that can automatically identify traffic flow is therefore necessary in order to help the authorities determine whether roads are busier or less congested. Developed nations have already created a system that Bangladesh is unable to afford. Therefore, I've made my choice to create a system that will assist the authority in detecting, tracking, and obtaining traffic flow at a reasonable cost. In order to create a system to recognize and track vehicles at a minimal cost, I have employed transfer learning of convolutional neural networks. Transfer learning is a system where we may reuse the code. I used 2 convolutional neural network models(CNN), 1 algorithm, and transfer learning techniques throughout. They are YOLOv8, and YOLOv7. The entire model's mAP has been produced. The YOLOv8 model, however, provides the highest mAP of the two. In the future, I'll focus on obtaining data on how each vehicle on the road is affected by traffic flow and I'll update the video dataset to obtain more accurate data and mAP for traffic analysis.