Liu Zhi-fang

A fast coarse-to-fine vehicle logo detection and recognition method

This paper presents algorithms for vision-based classification of vehicles based on vehicle logo in monocular image of traffic scenes recorded by a stationary camera. The intended application is automatic recognition of vehicle type for secure access and traffic monitoring applications, a problem not hitherto considered at such a level of accuracy. In recognition of vehicle type, differing with other researchers who attend to the recognition of shape, size of vehicle, we pay attention to the location and recognition of vehicle logo. The vehicle logo is unique mark of vehicle type (both make and model). We demonstrate that a relatively simple vehicle logo recognition method from front images can be used to obtain high performance verification and recognition of vehicle type. Firstly, the vehicle logo can be rough detected by prior knowledge, such as license, and then logo can be exactly detected by edge feature. Finally, the logo can be recognized using template matching and edge orientation histograms. Experimental results show the effectiveness of the proposed method.

A Real-time Vision-based Vehicle Tracking and Traffic Surveillance

A vision-based detecting and tracking vehicle in video streams is an important research in computer vision, and it plays an important role in ITS. The aim of motion detection is to get the changed region from the background image in video sequences, and it is also very important for targets classification and tracking motion objects. In this paper, a self-adaptive background subtraction method for vehicle segmentation was proposed. To indicate motion mask regions in a scene, instead of determining the threshold value manually, we use an adaptive thresholding method to automatically choose the threshold value. In order to accurately locate vehicle, we combined the projection of the difference image and the projection of the edge map from coarse to refine accurately locate vehicles. This proposed method could locate vehicle well. We formed an association graph between the regions from the previous frame and the regions from the current frame, so we modeled the vehicle tracking problem as a problem of finding maximal weight association graph. Very promising experimental results are provided using real-time video sequences, Experimental results demonstrate the validity of the approach in term of robustness, accuracy and time responses.

Face detection and facial feature extraction in color image

Face detection and facial feature extraction plays an important role in video surveillance, human computer interaction and face recognition. Color is a useful piece of information in computer vision especially for skin detection. In this paper, we propose a novel approach for skin segmentation and facial feature extraction. The proposed skin segmentation is a method for integrating the chrominance components of nonlinear YCrCb color model. The goal of skin detection is to group pixels to form possible face candidate regions and then use connected components analysis for pixels grouping. In order to detect the facial feature in scale invariant, the possible face candidate regions will be normalized, and then texture information in these regions will be segmented by means of mean and variance of face region. Edge will be detected using the method based on multi-scale morphological. Eye will be located by the PCA edge direction. The others feature, such as nose and mouth, also located using the geometrical shape information. As all the above-mentioned techniques are simple and efficient, the proposed method is computational effective and suitable for practical applications. In our experiments, the proposed method has been successfully evaluated using two different test datasets. The detection accuracy is around 98%, the average run time ranged from 0.1-0.3 sec per frame.

Change detection based on DSM and image features in urban areas

On the basis of stereo image analysis, the change detection of man-made objects in urban areas is introduced. Information of the height of man-made objects can be applied to reinforce their change detection. By comparison between the new and old DSMs, the changed regions are extracted. However, our aim is to detect changes of man-made objects in urban area and further in the potential areas by the means of line-feature matching and gradient direction histogram. The experiments based on the aerial images from Japan have proven that the algorithm is correct and efficient.

HOUSE CHANGE DETECTION BASED ON DSM OF AERIAL IMAGE IN URBAN AREA

The change of house often brings on the change of DSM in an area over different periods. If we can apply information of the height of houses to reinforce the house change detection, the reliability and efficiency of detection methods will be improved greatly. From this viewpoint, a new approach taking advantage of both height data from an image pair and image data is proposed to detect the house change in urban area and is called “data fusion technology”.

A novel face recognition method based on 3D face model

A new algorithm that project 3D face model to 2D planes is proposed. 3D face models are projected to different directions and a series of 2D face images are obtained. The projected 2D results are used as templates to match the input face images with different poses. The method that constructs 3D face model with 2.5D scans acquired by Minolta Vivid 910 is studied and a 3D face database containing 10 persons is built. Experimental results show that the proposed algorithm improves the recognition rate greatly for face images with pose variations than the method that uses just frontal images as templates.