Saliency Detection And Its Application In Intelligent Transportation System

In image processing and computer vision,saliency detection is a hot research topic nowadays,and has been applied to such fields as target detection,target tracking,medical image processing and so on.In this paper,the existing saliency detection algorithms at home and abroad are studied in depth,and combined with traffic signs and vehicle detection methods.So saliency detection can apply to the intelligent transportation system.After considering the accuracy and real-time of the algorithm,a traffic sign and vehicle detection system based on saliency detection is proposed.The main work is as follows:(1)In traffic sign detection,the color of the traffic sign is unique,so we can use the saliency model based on the global color contrast to detect the traffic sign.In order to reduce the number of colors in the image to reduce the computational complexity and remove the noise,the image is quantitated and smoothed in color space.Then the image is divided into multiple regions by graph-based image segmentation,in which the traffic signs will be located in one region separately.Finally,the obtained saliency map is weighted and constrained in space and color to obtain the traffic sign area.(2)According to the difference between vehicles and roads,the paper chooses the saliency detection method-BMS algorithm,by which the background and foreground are divided,to separate the vehicle candidate regions.Firstly,a plurality of binary images are obtained by using adaptive thresholds on three channels of Lab space,and then the attention map is obtained from the binary map by the flood-filled algorithm.Then,the average attention map is obtained by weighting the attention map.At last,the vehicle candidate regions are obtained.(3)The image library is used to train feature clustering tree.Because the SURF algorithm is robust in both illumination and scale transformation,we select the SURF algorithm to extract image features.The multi-layer k-means clustering is used to get the feature clustering tree and the TF-IDF algorithm is used to determine the leaf node weights.The extracted candidate region feature is sent to the feature clustering tree and the most similar image is determined to determine whether the candidate region is a vehicle region.(4)The traffic signs and vehicle detection algorithms are applied to the system,and the system is implemented by modules.