Vehicle detection is a significant step for many applications such as automatic monitoring of vehicles, traffic control, transportation, etc. In these systems, a fixed camera is usually installed in some places such as highway, street and parking. Images captured by the camera firstly will be processed, and then the vehicles in it will be identified. Generally, in these systems, cars are the favourite’s parts in a video frame. Therefore, providing methods that can increase overall accuracy and also the efficiency in vehicle detection is very important. Several methods have been proposed by researchers to detect cars in images. Frame subtraction is one of the most known in which every change between the current frame and the background image is regarded as a moving car. Since car shadow is identified as variable pixels, so there is a connection between the shadow of a car and that car, which causes two problems in these systems: Firstly, the actual shape and appearance of the car can disappear because of this conjunction. As a result, it is difficult to detect the location of the vehicle and to segment the image based on cars. Secondly, the shadow of one or more cars may interfere, and all of them are identified as a vehicle. These items cause many problems in monitoring systems and traffic control systems such as car counting, or vehicle location estimates and behaviour analysis. So shadows need to be identified. To solve this problem, recently, Karami has developed a method based on the region growing to remove the shadow of the vehicle. First of all, we should find the seed point, and then, based on the Euclidean distance, we will consider the distance from the seed point to each pixel of the image in the feature space as a weight. We will use these weights to separate the shadow of the car. The main problems about that are the complexity and computational speed, low precision and also a strong dependency on grey-grade changes. Therefore, in this research, we want to solve the above problems by improving the method of Background subtraction by weighting the pixels of the image by combining several texture features to remove the shadows of vehicles. To do this, each pixel in the background image and the current frame is weighted based on a combination of textural features. This makes the shadows and the background (asphalt) pixels to have very close values and thus removed in subtraction. The proposed method is evaluated on four datasets based on OA, HR, FAR, MODP and MOTP criteria.
Using these criteria, the proposed method is compared and evaluated with region growing method, median and averaging methods, and several other methods in shadow detection.
However, in general, according to the results, the proposed method outperforms than other algorithms by 2 to 15 percent improvement in mentioned criteria. To continue and complete the research in the field of vehicles detection and remove of shadows recommended:
- Using several different data, which are obtained in different situations and at different times.
- 2. Using other non-statistical colour and texture features.
- Detection of vehicles based on geometric structure