Research On Intelligent Recognition Algorithm Of Parking Space Based On Monitoring Image

Since the beginning of 1980 s, automobile industry is developing more and more quickly. The traffic problems are becoming more and more seriously. In order to solve the increasingly prominent traffic problems, Europe and America successfully established intelligent transportation system(ITS). It mainly combined with a variety of techniques such as electronic sensor, computer science. The recognition system of parking space is an important part of ITS. Primitively, scholars at home and abroad have proposed many kinds of methods to identify parking space intelligently based on sensor. They mainly are realized by using the induction coil and infrared, microwave, ultrasonic etc. But these methods based on sensor can be affected by the external flow, temperature and so on. Especially a sensor can only detect one parking space at a time. In recent years, scholars start to study recognition algorithm based on image to recognize parking space intelligently.The recognition algorithm proposed in this paper is mainly designed to identify indoor parking spaces. It analyzes occupancy of parking space through existing monitoring camera. In the process of implementation, the recognition algorithm based on image need to solve three problems. Firstly, how can we extract the interested parking space areas from the image? Secondly, after extracting parking space areas, what kinds of features information should be extracted from the parking image for recognition? Finally, according to the extracted features, which kind model should be established to identify the state of parking space? According to these three problems, the recognition system of parking space is divided into five parts: coordinates calibration, target binarization, shadow detection, occlusion recognition, and classifier selection.In the process of implementation, the prominent contribution in this paper has three aspects. Firstly, this paper adopts an improved method for shadow detection based on edge feature. The purpose is to eliminate the interference of shadow. The main improvements are as follows. In the process to eliminate the background texture, this paper considers the spatial information of pixels. Especially this paper finds the four valleys of the horizontal and vertical projection curve as coordinates of parking spaces. Compared with the traditional algorithm, this method can be better applied to the image which’s pixels are low and the boundaries are not obvious. Secondly, this paper applies an improved algorithm based on decomposed 3D Otsu to detect the target. The purpose is to keep the target point information more. The main improvements are as follows. This paper changes the first dimension of 3D Otsu from the original gray value into gradient value. The purpose is to enhance the edge information. This paper regards these points whose gray value is less than 20 as the background. Compared with the 1D Otsu and the decomposed 3D Otsu, this paper can effectively eliminate the noise disturbance, and can better retain edge information. Finally, this paper utilizes a kind of occlusion recognition model according to the distribution of occlusion. Each parking space is unevenly divided into n ine blocks. Each part is assigned different weigh. The four corners are assigned the smaller weights. The value is 0. The intermediate region is assigned the greater weight. The value is 2. The reason is that corners contain less information of vehicle, but intermediate region contains more information. The final average recognition rate is 97.59%. The false positive rate is 0.47%. The false negative rate is 1.94%.