A smart sensor network for object detection, classification and recognition

Abstract

Summary form only given. A surveillance system is a closed-circuit television system used to maintain close observation of a person or group. It is widely used nowadays to help a guard with consecutive sensing information. However, the concurrent observation of several monitors and the long-term exhausting visualization cause the problem of decaying attention. Furthermore, two major issues in the traditional surveillance systems affect their performance. First, the object resolutions are changed due to the varying distances between the object and the camera. Therefore, it causes the problem in object recognition since we have to adopt different sizes of the mask to properly extract object features. The other issue is that the detection of the moving object becomes difficult when the camera is not fixed. Consequently, the development of an efficient, automated surveillance system is an important task to overcome the problem for ensuring robust security. In this presentation, a sensor-based surveillance system is developed for object detection, classification and recognition to provide a more powerful and reliable system than the traditional ones. The wireless sensors are utilized as the guarders by only utilizing limited functions to detect the coordinates of the unauthorized invasions. When there are no specific signals detected by the wireless sensors, our smart sensor network system periodically rotates as traditional surveillance for safeguard. Once the wireless sensors detect any unauthorized invasion, the system adjusts the cameras toward the suspicious area and obtain the coarse image features for object classification. In order to reduce unnecessary processing and perform the surveillance system efficiently, a hierarchical feature extraction approach is adopted. That is, after receiving vigilant signals from wireless sensors, the cameras extracts coarse image features for object classification. Once the classification result shows that the object is a dangerous intrusion, the cameras continuously extracts fine image features for object recognition. Experimental results shows our system not only efficiently detect any unauthorized invasion, but also successfully classify and recognize the invasion.