基于高点视频的交通事件检测技术

Traffic Incident Detection Technology Based on Highly-shot Video

与传统近距视频不同的高点视频检测范围大、空间尺度变化大、场景复杂度高、交通事件种类繁多。结合传统近距视频检测技术,提出了一种基于高点视频的交通事件检测技术。首先利用混合高斯模型进行了背景建模及对象级的背景更新,实现了运动目标的检测。再利用基于区域预测和结合特征的方法进行了运动目标跟踪,最后采取基于车辆目标特征与区域职能的独立、联合事件检测模式实现了高点视频的交通事件检测。该方法实现了基于高点视频的交通事件检测,提高了对复杂大场景下高点视频的可解读性。

Compared with traditional close-up video, highly-shot video can detect larger range of scene, it contains larger-scale changes in space, a higher complexity of scene, and more different types of traffic incident. We proposed a method of highly-shot video based traffic incident detection technology on the basis of learning from traditional close-up video based detection. First, we used Gaussian mixture model to perform background modeling and object-level background update for moving target detection. Then, we combined region-prediction-based and feature-based methods to track moving targets, lastly, by using vehicle features and regional independent and joint incidents detection to finish the traffic incident detection. We implemented the method to realize the highly-mounted video-based traffic incident detection, and increased the interpretability of highly-shot video with large scenes.