Vehicle Detections and Speed Estimation From Roadside Video Using Deep Learning Approach

Driving on the busy urban road is not easy, with the highways being crowded by vehicles. Information about the road's environment is crucial, and closed-circuit television (CCTV) cameras are fully deployed on national highways for road environment analysis and surveillance. This study demonstrates vehicle detection vehicle detection and speed estimation without using any sensors. We extract roadside video using CCTV cameras. YoloV8 and DeepSORT use deep learning approaches for object detection and tracking. The research illustrates the overall analysis of deep learning model implementation on different roadside environments, i.e., day, night, and rain, with different CCTV camera angles on both the left and right sides of highways. Our study classifies vehicles into three categories: car, bus, and truck, and the evaluation of object detection models implemented is based on precision and recall. For the car, we got precision, and the recall value is 98%; for the truck, we got 97% and 90%. The bus's precision is 60%, and the recall is 100%. MAPE and MAE are implemented to evaluate speed calculation results, and we find the significant factors that impact the speed calculations.