| Summary: | 碩士 === 中原大學 === 電子工程研究所 === 106 === The surveillance systems have been widely used in our daily life. The functions of current surveillance systems simply record images, they cannot autonomous determine and referee the pedestrians. For these purposes, we use the manual monitoring. Focusing on the screen monitoring causes the fatigue of people and reduces the overall efficiency. This study proposes the methods of detection and tracking for pedestrians. When a pedestrian passes through the range of photography of the surveillance system, the system will automatically perform the detection and path tracking. The systems provide the target numbers and calculate the amount of all pedestrians in the area. The information of the detection and tracking of the pedestrians can be used by the administration.
This thesis is divided into two developed systems: detection and tracking. We use the KNN-based background subtraction algorithm to obtain more accurate images of moving objects. And we adopt the morphological techniques to filter out noise and cut off the shadow. Compared with the histogram of oriented gradient (HOG) pedestrian feature, we can recognize the moving object as a target. For the tracking system, when it is confirmed that the moving object is a pedestrian, the system will give a number and counter the amount. It will not be given the number repeatedly, and it can count the total number of pedestrians in the recognized area. Finally, the system will judge whether the pedestrians leave the detection range. The system will remove the tracking frame and the tag number for the pedestrians leaving the detection range. The design methods proposed in this thesis can achieve a correct judgment rate of 86% in the test film, and the processed speed is 6.5 frames per second.
|
|---|
