Study Of The Vehicle Detection Method Based On Parts In Cooperative Vehicle Infrastructure System (CVIS)

The vehicle detection is one of the key technologies of the Cooperative Vehicle Infrastructure System(CVIS). As an important research field of intelligent transportation, it has received much attention from scholars. Though the detection method based on machine vision has achieved remarkable results, there are still some problems due to complex traffic environment, weather changes and other effects. In this paper, the vehicle in traffic scene is studied, and the detection method is based on parts.According to the camera perspective projection imaging principle, size and shape change in 2D image, through the traffic scene calibration, inverse the 2D image projected onto the 3D space, and set up the inverse projection surface to reconstruct the data of inverse projection. It can avoid 2D image problems effectively. The vehicle target can be characterized by multiple parts, and through the combination of parts to complete the detection of the target. The vehicle target shows different characteristics at night and during the day. The front brightness of headlights at night is very significant, while the color of vehicle taillight and license plate during the day is very obvious. The former divides target by background subtraction, then extracts the target light parts and pairs target by combining lights geometrical characteristics. At last, Gaussian mixture model is used to complete the accurate detection of target combining scale factor. The latter locates rear lights and license plate parts by the color conversion model through spatial relationship between the parts, with the Markov Random Field, and thus the vehicle detection is achieved.The proposed method being effectively used in CVIS and the results show that the vehicle target can be better detected. In the daytime scene, the accuracy detection rate is 93%, and at night it is 94.5%. This method has good real-time performance.