Research On Key Technology Of Person Re- Identification For Monitoring Image

Person re-recognition has important applications in many fields such as intelligent surveillance, human-computer interaction, robotics and so on. It has become a hot research topic in the field of computer vision and pattern recognition. However, it is challenging to study person re-recognition under various factors, such as illumination, background, angle and posture.Current person re-recognition methods mainly include the traditional manual design model based methods and the deep learning based methods. The traditional methods include manual feature detection and distance measure, and one of their main problems is the manual designed features will face the difficulties of satisfying the complicated and changeable environment. Deep learning methods show their advantage of automatic feature extraction and are used for person re-recognition quickly, but they have to face the problem of lack of training samples. Furthermore, both the traditional methods and the deep learning based methods consider only the effectiveness of person re-recognition, and thus will result in a big gap between the experimental results and the actual results.To solve above problems, an deep learning based End-to-End person re-recognition system is studied and implemented in this thesis. Firstly, a Faster R-CNN framework with the combination of RPN and R-CNN is used to detect the person of any size in the surveillance image; Secondly, the deep convolutional networks model is adopted to determine whether the persons in the object image and the detection image are same persons, with the convolutional network model of symmetric structure and sharing weight for feature extraction, and by transfoming the multi-classification problem into a two classification problem for person re-recognition. Finally, an End-to-End person re-recognition system is realized based on the above technology.Simulation results verify the effectiveness of the proposed system?...