Integrated Vehicle Detection and Noise Removal in Traffic Footage Using CNN, Kalman Filter, and Canny Edge Detection

Abstract

Computing methods are being researched more and more for automatic traffic surveillance due to the growing requirement for effective traffic control and monitoring. Reliable vehicle recognition, counting purposes, and noise reduction from such material are essential for studying traffic, development, and roadway surveillance in densely populated contexts. The suggested approach includes using Canny edge detection for picture segmentation, a filter called Kalman in noise reduction, and Convolutional Neural Networks (CNN) for vehicular recognition. We study the potential synergies between these methods to improve vehicle identification performance in difficult traffic situations regarding precision, sensibility, and specificity. The successful use of the suggested comprehensive strategy is demonstrated through a demonstration employing Vehicle Recognition and Counting Employing YOLOv8 and Byte Tracker database from Kaggle. Evaluate the reliability of vehicle identification findings by examining indicators of success such as specificity, sensitivity, and preciseness. According to this research, the entire approach works better than the separate methods, obtaining higher results for precise counting of vehicles and identification while also successfully lowering noisy abnormalities in the video clips. From the results obtained, the proposed MLP produces an Accuracy of 94%, a sensitivity of 0.90, and a specificity of 0.91. The tool used is Jupyter Notebook, and the language used is python