Dynamic Attributes Based Target Detection And Recognition

Digital image processing technology, which is one of the emerging fields of scientific research, implies great potential for development. Compared to traditional detection methods, video-based traffic information detection has the advantages, such as: simple installation, easy maintenance, large detection range and information-rich. General methods for video based target detection and recognition are based on targets’ dynamic attributes, static attributes or the fusion of both attributes.Common ways for video based detection and recognition are mainly through monocular based, stereo vision based and fusion of active sensor-vision based methods. This article is focusing on the dynamic attributes of target, discussing the application of monocular, stereo vision in different scenes. The first scene is concerned with the public transportation road through monocular based method. The moving vehicles and pedestrian are considered as the targets of detection and recognition. The vehicles’ motion trajectory fields are proposed to detect suspicious target behavior, such as vehicles’ unauthorized turns, pedestrian crossing the road. The second one, using stereo vision based method, is according to the inner scene of public bus. This article considers the passengers as detection and recognition target. The interference trajectories created by false targets should be distinguished by using na?ve Bayes classification method to improve the passenger flow counting accuracy.This article, using the video scenes captured by camera, considers the motion objects as research target. The methods under the two scenes are almost sharing the same steps: first the method obtained the trajectories by tracking moving target, then extracted the dynamic features from each trajectory, and finally completed the target detection and recognition based on learning motion features of normal motion targets.The methods according to each scene have been tested by multiple video samples respectively. The results showed that target detection and recognition by using its motion features satisfied the real-time requirement and high efficiency of detection and recognition.