نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشکده مهندسی برق، دانشگاه صنعتی سهند، تبریز، ایران
2 دانشگاه صنعتی سهند
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Human action recognition is one of the most important applications of image processing which has gained great attraction in computer vision. Because of the complexity in image and video analysis, state-of-the-art proposed algorithms are still far from developing error-free and fully generalized systems. This work presents a new method based on key-poses of frame silhouettes for the task of two-person interaction recognition. In the proposed algorithm, an IMTF-based shape descriptor which gives a perfect description of the shape due to its ability to collect data from the whole shape. The task is useful for offline or online applications such as video content analysis and video surveillance. The main idea is to perform a two-step classification based on a sequential pattern mining classifier. So, we extract the shape of the foreground silhouette of the persons and describe it using the invariant multiscale triangle feature to compare each frame with a pre-defined dictionary of key-poses. Then, each frame is labeled as one of the existing classes. The output of this step is a sequence of labels which is the input of a sequential pattern classifier. The obtained results (93.9% for SBU database, 92.4% for UT database, 86.6% for BIT dataset and 86.9% for NTU RGB+D dataset) indicate the accuracy and efficiency of the proposed method and provide an attractive solution to the problem of sequence classification.
کلیدواژهها [English]
)