Reseach On Person Attributes Recognition Under Video Surveillance

Object recognition is a hot topic in the field of computer vision, and its main purpose is to identify interested targets from complex image scenes. Moving object recognition is the key factor to intelligent monitoring system. It can distinguish the person, vehicle, animal, etc. Particularly, the person recognition is one of the most important topics in moving object recognition research.However, due to the technical limitation, ordinary image analysis technologies can only distinguish pedestrian object from multiple moving objects but cannot achieve further analysis and identifications. Person analysis will provide great amount of attribute information to the monitoring system, which can be applied into abnormal event detection, such as person re-identification, the mall crowds statistical analysis, and other fields. However, due to the cost limitation, most agencies equipment with low resolution image monitoring devices, which are not qualified to provide clear enough images. Therefore, it is hard to be applied into real monitoring systems based on ordinary image analysis technologies. With consideration of the difficulty, this paper focuses on identifications of person’s characteristics differentiated with scenarios. The main contributions of this paper state as follows:1. There are two aspects inside target tracking problems. One is a proposed target tracking structure based on the idea of the three-step search algorithm. This method can effectively track the target, and the algorithm time-consuming is low. It is able to achieve real-time in the monitoring system. Another aspect is an improved particle filter tracking algorithm. The number of particles needed to track the target has been effectively reduced. It is able to reduce the tracking time-consuming.2. This paper proposes a recognition algorithm for pedestrian movement. For the different states of motion, such as walking, running, cycling, a hierarchical recognition algorithm is proposed for analysis. By analyzing the different local features of pedestrians in different motion, and extracting effective localized-features, combining with gait recognition technique, a stepped motion-state recognition can be achieved.3. This paper proposes pedestrian gender recognition algorithm according to the appearances. For the low pedestrian resolution problem in the monitoring scenarios, this paper proposes a calculation frame for regional-based pedestrian gender recognition algorithm. This gender recognition algorithm can recognize pedestrian genders through integrated rounds of recognitions based on different features. In addition, this paper also uses the algorithm framework to solve particular occupation recognition problems.4. This paper creates a detection algorithm for pedestrians carrying parcels. In order to provide a richer character information to the monitoring system, the detection algorithm is proposed to detect the pedestrians’ carried packages. Hitherto, some preliminary detection accomplishments have been achieved.