Training Object Detection Algorithms for Vehicle Type Classification from Public CCTV Camera by Using Custom Dataset

Computer vision technology has gained popularity and been widely applied across various fields. Object detection algorithms are pivotal in these applications. Traditional algorithms have been developed to classify over 80 general object types. However, they still fall short of comprehensive object classification in specialized domains. This research presents the training of object detection algorithms to classify vehicle types using a custom dataset from collected data from multiple public CCTV cameras. Performance evaluation results demonstrate that Precision, Mean Average Precision (mAP@50), mAP@50–95, Recall and F1-Score achieve values 0.83, 0.72, 0.58, 0.58, and 0.69 respectively. Testing results indicate that the proposed algorithm achieves 93% higher accuracy in classifying vehicle types compared to traditional algorithms. This algorithm can be further refined and applied to traffic management systems to maximize future benefits.