real time traffic analysis

this model estimates vehicle speed by leveraging object detection, tracking, and perspective transformation. it begins by using yolov8 to detect vehicles in each frame of a video. yolo provides bounding boxes around detected objects, along with confidence scores and class labels, allowing the model to differentiate between cars and trucks. vehicles classified as cars are displayed with peach-colored annotations, while trucks and semis are shown in green. once a vehicle is detected, the model assigns it a unique tracker id using bytetrack, ensuring consistent identification even as the vehicle moves across multiple frames.

to calculate speed, the model records the vertical position of each tracked vehicle over time. since cameras often capture footage from an elevated or angled perspective, the raw pixel displacement of a vehicle may not directly correlate to real-world movement. to address this, the model applies a perspective transformation using a predefined mapping between a source polygon (representing the detected region in the original frame) and a target rectangle (representing a standardized measurement plane). this transformation ensures that movements in the frame correspond proportionally to real-world distances, allowing the model to calculate speed accurately regardless of the camera angle.

the model maintains a history of tracked positions for each vehicle, storing them in a time-sequenced queue. when a vehicle has been tracked for long enough, the system calculates its displacement over a known time interval by comparing its initial and final transformed positions. since the frame rate of the video is known, the time elapsed between position updates can be precisely determined. the model then applies a simple kinematic formula, converting the displacement and time into speed, expressed in kilometers per hour.

this approach allows the system to work across different camera angles because the transformation matrix is adjustable. by defining a new source-target mapping specific to each camera setup, the model can adapt to various perspectives while maintaining accurate speed estimation. additionally, because vehicle classification is handled by yolo’s object detection network, it remains robust to changes in lighting, background, and road conditions, ensuring reliable classification of different vehicle types in diverse environments.