Image segmentation and shape feature extraction methods for vehicle classification in thermal video sequences

Abstract

Recently, due to a drastic improvement in image processing and computation technologies, automatic target detection, segmentation, and classification methods have been studied plentifully in object recognition areas. However, since thermal images represent only the temperature difference between objects and background and they have more blurred edges than color images, it is hard to segment objects correctly in them and the segmented images have noisy object boundaries. Therefore, automatic target segmentation and classification in thermal images are difficult task in object recognition area. In this thesis, to overcome these problems, an image segmentation method and a new shape feature extraction method are proposed for vehicle classification in thermal images. To segment the target vehicles correctly in thermal video sequences, a spatiotemporal parameter update method for ViBe is proposed. The proposed change detection method was tested and evaluated with various sequences, demonstrating that it outperforms state-of-the-art methods. In addition, to improve the vehicle classification performance in thermal video sequences a novel feature extraction method based on the Target Trait Context (TTC) is proposed. The keypoint repeatability test and the classification performance test were performed and compared with those from the previous methods. Experiment results show that the proposed feature works well for thermal video sequences and outperforms the previous methods in classification performance.