Application of YOLOv8 Small Object Detection in Traffic Flow Congestion Prediction Models

This study aims to analyze traffic surveillance videos at four key points of the Changshen Expressway by integrating multiple algorithms to accurately detect moving vehicles and quantify traffic flow. Initially, the ViBe background modeling algorithm is employed for dynamic target recognition, followed by the YOLOv8 algorithm for traffic flow statistics. The ViBe algorithm rapidly updates background information through a pixel-based background subtraction model, suitable for handling complex environments such as changes in lighting and background noise, effectively performing vehicle detection and foreground separation. The YOLOv8 algorithm achieves object detection and classification in a single neural network inference, rapidly and accurately locating and identifying multiple moving targets in traffic scenes, making it suitable for vehicle identification and classification tasks in high-frequency traffic flow monitoring. By combining these two algorithms, traffic flow is collected once per second, generating key traffic parameters including vehicle density, speed, and lane occupancy rate, providing data support for subsequent traffic optimization and emergency lane management. An LSTM neural network is used to establish a traffic congestion prediction model, predicting the traffic conditions between the third and fourth observation points every 10 minutes, updating the congestion status in real-time. Data analysis reveals traffic dynamics within the first 500 seconds at different observation points during peak hours, highlighting traffic conditions. The LSTM model's evaluation metrics include MAE, MSE, MAPE, and RMSE, providing important quantitative evidence for the model's performance.