Design and Implementation of Pedestrian Detection System for Vehicle Application

Autor: Jia-Hua Wu, 吳佳樺
Rok vydání: 2015
Druh dokumentu: 學位論文 ; thesis
Popis: 103
Many automakers and drivers put more emphasis on driving safety in recent years, wherein the pedestrian detection is one of the key functions. The pedestrian detection systems applied by the automakers are often based on dual sensors, such as the combination of the radar and the cameras. This paper only applies single camera as the sensor, as well as combines the image processing technology and the Support Vector Machine (SVM) to develop the algorithm, which is low-cost and can be applied to all kinds of cars. Many previous researches applied the SVM classifiers to design the pedestrian detection systems. However, complex and varied real driving images resulted in low pedestrian detection rate and high error rate. This paper proposes a pedestrian detection system for vehicle applications, which applies the image processing technology to obtain the pedestrian candidate blocks. During the process of pedestrian recognition, the color blocks of pedestrian’s body are used as the detection unit, and the weight screening of the upper body is applied to detect or adjust the candidate blocks. Finally, the possible pedestrian candidate blocks should go through the SVM detection, which filters many candidate blocks that do not need detection and improves the overall system performance. In addition, this paper divides the SVM detection into two-stage detection of the upper body and the lower body, which can obtain higher recognition rate, and enhances the driving safety. The final results show that the proposed pedestrian detection system can be applied to each weather and different scenes. The average detection rate is higher than 94%, and the false-positive rate is no more than 5%, which shows that the proposed system can achieve certain pedestrian detection during real driving.
Databáze: Networked Digital Library of Theses & Dissertations
Externí odkaz: