基于多视点空间目标的高原公路交通行为检测

Detection of Highway Traffic Behavior in Plateau Based on Multi-View Spatial Target

为减少高原公路交通事故发生几率,提出一种基于多视点空间目标的高原公路交通行为检测方法。建立双轴旋转以经纬度划分区域的多视点投影模型,利用组合不变矩灰度变换目标车辆彩色图像,采用BP神经网络中的Fletcher-Reeves共轭梯度法对视频图像去抖,完成图像预处理,提取目标车辆图像中最需要的目标。依据提取目标,采用直方图均衡化算法来对图像进行增强处理,将车辆最终的增强结果采用多视点空间目标与最小外接矩形结合检测的形式给出,完成交通行为检测。实验证明:所提方法对高原公路交通行为识别正确率较高,识别高原交通行为具备高精度,不受外界因素干扰,对高原公路车辆的交通行为作出提前预判,可降低车辆交通事故发生几率。

In order to reduce the probability of highway traffic accidents in plateau, this paper presented a method to detect highway traffic behavior in plateau based on multi-view spatial targets. The multi-view projection model with biaxial rotation was built, and longitude and latitude were used to divide the area. The color image of target vehicle was transformed by the gray level based on combined invariant moment. The Fletcher-Reeves conjugate gradient method in BP neural network was used to remove the image dithering, so that the image preprocessing was finished. Moreover, the most needed target in vehicle image was extracted. According to the extracted target, the histogram equalization algorithm was used to enhance the image. Finally, the enhancement result of vehicle was given in the form of combing multi-view space target with minimum external rectangle detection, and thus to complete the traffic behavior detection. Experimental results show that the proposed method has high recognition accuracy for highway traffic behavior in plateau, high identifying accuracy for the high-altitude traffic behavior, and it is not affected by external factors. In addition, this method can prejudge the traffic behavior of vehicle, so that the probability of traffic accident can be reduced.