Publicação
Human features detection in video surveillance
| Resumo: | Human activity recognition algorithms have been studied actively from decades using a sequence of 2D and 3D images from a video surveillance. This new surveillance solutions and the areas of image processing and analysis have been receiving special attention and interest from the scientific community. Thus, it became possible to witness the appearance of new video compression techniques, the transmission of audio and video in real-time, targeting identification and tracking objects in with complex environments. Traffic monitoring, automotive safety, people counting and activity recognition applications are examples. With the development of sensors, new opportunities arose to expand and advance this field. This dissertation presents an activity recognition system to recognize human motion. The system does not need optical markers or motion sensors. This human activity recognition system is divided in three stages: human segmentation, in an outside and inside environment; extraction of the human features; and classification models to detect the human actions. Therefore, the main objective in this work is to develop an algorithm to extract human features. This algorithm aims to develop a new representation and extraction method using a sequence of features in a skeleton silhouette. Mainly, the segmentation of humans is based on a previous work, centered on background subtraction. An algorithm is applied to convert the object captured in the video surveillance to a binary image using a skeleton algorithm. Afterwards, and based on the physical parameters of the human motion, it becomes possible to discover the principal features of the human skeleton, called physical features, head, hands and feet. The viability of using features detection in a human recognition system was tested and compared with other existing systems. The results point out that the system has good performance (8.96% of perfect match and the average rate was 68.65%). Nevertheless, in images where the features of the human body are covered, with umbrella or heavy coats for example, the system presents certain limitations. This process has a high execution speed and a low cost computational processing: average of 5910 µs with a standard deviation of 5650 µs. In the near future, classification models to detect the human actions will be included. |
|---|---|
| Autores principais: | Barbosa, Patrícia Margarida Silva de Castro Neves |
| Assunto: | Video surveillance Human features selection Skeleton Vídeo vigilância Características humanas Esqueleto |
| Ano: | 2016 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Human activity recognition algorithms have been studied actively from decades using a sequence of 2D and 3D images from a video surveillance. This new surveillance solutions and the areas of image processing and analysis have been receiving special attention and interest from the scientific community. Thus, it became possible to witness the appearance of new video compression techniques, the transmission of audio and video in real-time, targeting identification and tracking objects in with complex environments. Traffic monitoring, automotive safety, people counting and activity recognition applications are examples. With the development of sensors, new opportunities arose to expand and advance this field. This dissertation presents an activity recognition system to recognize human motion. The system does not need optical markers or motion sensors. This human activity recognition system is divided in three stages: human segmentation, in an outside and inside environment; extraction of the human features; and classification models to detect the human actions. Therefore, the main objective in this work is to develop an algorithm to extract human features. This algorithm aims to develop a new representation and extraction method using a sequence of features in a skeleton silhouette. Mainly, the segmentation of humans is based on a previous work, centered on background subtraction. An algorithm is applied to convert the object captured in the video surveillance to a binary image using a skeleton algorithm. Afterwards, and based on the physical parameters of the human motion, it becomes possible to discover the principal features of the human skeleton, called physical features, head, hands and feet. The viability of using features detection in a human recognition system was tested and compared with other existing systems. The results point out that the system has good performance (8.96% of perfect match and the average rate was 68.65%). Nevertheless, in images where the features of the human body are covered, with umbrella or heavy coats for example, the system presents certain limitations. This process has a high execution speed and a low cost computational processing: average of 5910 µs with a standard deviation of 5650 µs. In the near future, classification models to detect the human actions will be included. |
|---|