联合深度与彩色信息的对象检测研究-电子与通信工程专业论文.docxVIP

ABSTRACT Object detection is one of the key technologies in the field of computer vision and pattern recognition. It is widely used in many fields, such as industry, traffic security, military security, public places and artificial intelligence. More and more experts and scholars do research in object detection to develop new method to improve the quality of object detection. The current detection methods are mainly based on color images. Authors usually improve the features or the classifiers to get the high rate of the accurate. Color images provide rich color information, and helps make the detection more accurate through extracted color features, texture features, edge features, etc. However, color images have some disadvantages. They are vulnerable to illumination changes which may make a great impact on the extracted features. With the development of depth images, object detection technology experiences a new breakthrough. Depth information records the distance between the object and the camera. It is robust to illumination changes and can help to detect the edges and the occlusions. In this thesis, we use the combined depth and color information to detect objects. For human detection, various body poses, different clothing and some other factors may make the task difficult. We have done some researches on feature extraction, training classifiers, and used canny edge detection method to verify the candidates of detection results. In order to overcome the low quality of depth map captured by Kinect, we proposed a new method that uses different models to detect different separated parts. The experiment results show that our method improve the detection rate. For object detection, they don’t have the variable attitudes like humans. We use view-based objects from video sequences to extract the combined color and depth histogram of oriented gradients features and histogram of depth features to detect objects. Then we use support vector machine to learn the new mod