Due to aisle, emergency lanes and other functional areas with rare events and lack of monitoring, an algorithm for smart detection of vehicle in illegal parking area was proposed. Firstly, region of interesting was selected. Secondly, Gaussian Mixture Model was used for background construction, and moving objects were detected by background subtraction. Thirdly, morphological processing was used for detection of relatively complete target, then fusing of ellipse fitting, compacted features detection and tire circularity detection for vehicle recognition. Experiments show that this algorithm could realize monitoring exactly and keep illegal parking area smooth.