Data Processing and Augmentation Techniques for Vehicle Counting and Detection in the Philippines

Manual vehicle counting in the Philip-pines is challenged by rapid urbanization, while artificial intelligence (AI) offers a solution, hindered by the lack of representative vehicle datasets. This study addresses the dataset gap in vehicle detection by applying data engineering techniques to improve quality and mitigate class imbalance. Three methods were tested: grayscaling, blocking overrepresented classes, and grouping classes with similar instance counts. Grayscaling notably improved night-time detection, boosting precision and recall by 247% and 155%, respectively. Grouping cars and motorcycles yielded significant gains in both day and night settings, with day-time precision and recall increasing by 351% and 279%, and night-time by 157% and 164%. These results highlight the vital role of data preprocessing and balanced class distribution in optimizing AI performance for traffic management.