Abstract: A novel algorithm is presented to estimate real-time traffic speed by using images from uncalibrated PTZ cameras.Firstly,line parts of lane boundaries are obtained from the roadway-background images,and the camera is calibrated by estimating the slopes of such parts and their corresponding vanishing poin.t Then a generic lane boundaries detection via Kluge circular-model is carried out by Randomized Hough Transform.With the obtained lane boundaries,a vehicle tracker is established to track the moving vehicles between frames of the video sequences by a combination of the adaptive background subtraction and the extended Kalman filter technique.Finally,the vehicles'speed is estimated from images by transforming their image coordinates into the real-world coordinates by our simplified camera model.The experimental results show that,with great flexibility in camera calibration,our algorithm is both robust and efficient for the vehicle speed estimation.