The field of transportation management is seeing a rise in the importance of intelligent transportation detection and counting. Yet, detecting them remains difficult because of the wide range in vehicle sizes, which has a direct impact on the precision of vehicle counts. In this research, we present a vehicle counting and detection system that uses computer vision to solve this problem. The Intelligent Transportation System (ITS) has been developed through the enhanced optimization tool which belongs to the conventional time prediction framework, normally used in industry. The non-parametric approach has been developed with the help of machine learning approach and enhances the computational power. Huge historical information is used to generate the accuracy through traffic pattern with non-parametric approach. Additionally, this approach needs strong dataset, system for training and testing. The main problem with the current concept is that it takes too long to detect vehicles using machine vision. In order to alleviate traffic and safety concerns, the proposed model centres on precise vehicle recognition for usage in the intelligent transport system. The core components of the technique, which include convolution layers, pooling layers, and fully-connection layers, provide it exceptional robustness and allow it to efficiently complete the task of vehicle detection. The proposed pooling based dense- You Only Look Once (YOLO) model is constructed to enhance the accuracy in vehicle detection and eliminating the vanishing gradient issue. The pooling strategy is used to pool and update the multi-scale region features for enhanced detection. The loss function is established with MSE and the performance analysis shows that the proposed technique is performed well on UA-DETRAC dataset.