A Hybrid Camera Technique for a Pedestrian Safety Support System at a Traffic Junction

十字路口发生了许多交通事故，主要是由于驾驶员对行人和车辆的疏忽造成的。为了防止行人发生交通事故，大多数提出的系统都是从一般城市街道或高速公路环境下拍摄的汽车视频中检测行人。本研究提出了一种混合技术，根据车辆是否行驶或停车等车辆情况，在单摄像头和立体摄像头之间切换图像。当自动驾驶汽车向右或向左转弯时，该技术可以检测十字路口的行人和汽车等移动物体，并通过分析被检测到的行人的行为向驾驶员发出警告。采用各种车载视频图像对所提技术的性能进行了检验。本文介绍了运动目标检测、行人分割、行人行为识别等技术，并对各技术进行了评价。

There are many traffic accidents occurred at an intersection, and there are mainly caused by a driver’ s oversight of pedestrians and cars. T o prevent pedestrians from trafficaccidents, most of the proposed systems detect pedestrians from a car video taken under general urban street or highway environment. This research proposes a hybrid technique employing switching images between single and stereo camera according to the vehicle situation such as whether or not the vehicle is driving or stopping. The technique detects moving objects such as pedestrians and cars at an intersection when the ego car turns right or left, and gives a warning to a driver by analyzing the detected pedestrian’ s behavior. The performance of the proposed technique was examined employing various vehicular video images. In this report, moving objects detection, pedestrian’ s segmentation, pedestrian’s behavior recognition and finally the evaluation of each technique are described.