Traffic management requires data about traffic to control the progression of traffic. Gathering continuous traffic stream data is genuinely simple, as a huge number of traffic camcorders all over the planet go about as sensors. However, it is challenging to utilize that data to process and control traffic stream. More seasoned techniques like guided loop detectors (ILDs), infrared detectors (IRDs), and laser sensors for recognizing vehicles have issues like a significant expense, proficiency, and trouble. The techniques utilized in this article are regression-based counting and detection-based counting. It additionally assesses the achievability of utilizing pre-prepared profound learning models like detection-based Faster RCNN, SSD, and YOLO. In this paper, we assess our model performance and its productivity in light of a custom dataset. Our outcomes show the adequacy of the joined strategy as far as precision contrasted with utilizing every technique independently.