Detecting the moving objects of special trajectories by fusing laser range scanners and video cameras

在这两年的项目中，取得了以下成果：1.基于激光扫描数据的运动目标检测、跟踪和分类算法研究分两个层次进行。1)在交叉口等存在多种运动目标的环境中，提出了一种利用激光扫描数据对运动目标进行联合检测、跟踪和分类的算法。2)在只有行人的环境下，为了准确检测出行人的行走模式，估计行走轨迹，定义了详细的步行模型，研究了密集拥挤环境下基于粒子滤波的行人跟踪算法。激光与视觉融合的算法。它包括以下几个子主题：1)图像属性的提取；2)同一人的匹配；3)基于激光和视频的跟踪。在密集拥挤情况下，仅基于激光数据的精确跟踪存在局限性。同样，基于激光的跟踪可以n…除了XY坐标序列之外，更多的是获得其他信息。开发了激光与视觉融合测量系统、图像属性提取算法、轨迹与图像属性匹配寻找相同运动目标的算法、激光与视觉融合跟踪算法。行为分析-寻找特殊的轨迹。1)步态参数提取与步态识别，2)个人行为分析，3)群体行为分析。定义了详细的步行模型，提取了双脚的步长、速度、加速度等步态参数，提出了基于上述步态参数的个人步态特征识别算法。另一方面，基于步态参数和轨迹参数对个人行走行为进行了研究。提出了一种将个人行为分为正常行为和异常行为的算法。此外，开发了从轨迹质量中寻找主流的算法，并将结果应用于先进的空调控制，并在仿真上进行了测试。为了获得真实的数据，进行了大量的现场实验。本项目中的所有算法都是在真实数据上开发和测试的，旨在实现真正的应用。较少

In this two years project, the following results are achieved.1. Algorithms for moving objects' detection, tracking and classification using laser scan data.The topic is studied in two levels. 1) In the environment such as an intersection, where different kinds of moving objects exist, an algorithm is developed for joint detection, tracking and classification of moving objects using laser scan data. 2) In the environment where there are pedestrians only, in order to precisely detected walking pattern, and estimate walking trajectories, a detailed walking model is defined, and algorithms of tracking pedestrians in dense crowded situations using particle filter are studied.2. Algorithms for fusing laser with vision. It includes the following sub-topics : 1) extracting image attribute; 2) matching for the same person, 3) laser and video based tracking.In the case of dense crowded situation, there are limitations in precise tracking based only on laser data. Also laser-based tracking can n … More ot obtain the other information except the sequence of XY coordinates. So that a measurement system by fusing laser with vision, an algorithm of extracting image attributes, an algorithm of matching on both trajectory and image attributes to look for the same moving objects, and a tracking by fusing both laser and vision are developed.3. Behavior analysis - looking for special trajectories. It includes the following sub-topics : 1) walking parameters extraction and gait recognition, 2) personal behavior analysis, 3) group behavior analysis.A detailed walking model is defined, the walking parameters such as stride, speed and acceleration of both feet are extracted, an algorithm is developed to recognize personal gait characteristics based on the above walking parameters. On the other hand, personal walking behavior is studied based on both gait and trajectory parameters. An algorithm is developed to divide the personal behaviors into normal and abnormal ones. In addition, algorithm is developed to find the main streams from the mass of trajectories, and an application of the results for advanced air-conditioning control is proposed and tested on simulation.4. A number of field experiments are conducted to acquire real data. All the algorithms in this project are developed and examined on the true data and aiming for true applications. Less