基于可能性分布合成的车辆提取方法

A Vehicle Detection Method Based on Fusion Theory of Possibility Distribution in LIDAR Data

为了综合机载激光扫描与测距系统(LIDAR)异源数据所能提供的车辆高度、强度及轮廓等信息，减少目标与特征对应关系不明确对车辆识别的影响，以取得更好的车辆识别效果，提出了基于可能性分布合成的LIDAR数据车辆提取方法。首先，定义了差值融合规则对LIDAR首次回波与末次回波图像进行融合，并依据高度特征和面积特征分离地物点并去除大面积联通区域，提取出车辆预识别区域；其次，提取二值图像中的区域长宽比特征，并结合强度图像与可见光图像构建区域强度比特征；最后，对两种特征分别构造可能性分布，并进行分布的合成，以实现异源数据的协同决策。实验证实该方法有效。

In order to obtain a better vehicle recognition result, this paper proposed a method based onthe synthesis of probability distribution of light detection and ranging (LIDAR) data to extract vehicle.By fusing the characteristics of the multi-sources of LIDAR data, and reducing effect caused by thecorresponding unclear relationship between objectives and features. Firstly, a difference fusion rule ofLIDAR first and last echo image was defined. The height and area feature were obtained and large areaUnicom areas were removed, the pre-recognition areas of the vehicle were extracted. Secondly, thevalue of ratio of length and width of pre-recognition areaswere extracted, and the regional strengthcharacteristics were constructed by combing the intensity image and visible images. Finally, thepossibility distributions of these two features were constructed. After fusion and making collaborativedecisions, results were obtained. Heterogeneous data proved that this method is effective.