基于模式识别与ST-MRF相结合的车辆检测方法

车辆检测技术的主要难点是在于解决车辆之间的遮挡，以及由于光照变化引起的车辆与其阴影之间的遮挡问题，这些问题将直接影响检测的精度。针对这个问题，在原ST-MRF方法上研究了基于模式识别与ST-MRF相结合的车辆检测方法。模式识别技术分割相互遮挡的2辆车之间的边界，并识别相互遮挡车辆的边缘间隙以及边界信息，模式识别结果反馈给ST-MRF算法，算法对相互遮挡车辆重新分配标号，优化处理并融合不完整的分割部分，确定单个车辆信息。路段车辆检测实验结果表明，在检测区域行驶的325辆车，用原始ST-MRF算法跟踪统计到的车辆数为258辆，成功率为79%，采用模式识别技术与ST-MRF相结合算法统计到车辆315辆，成功率为97%；交叉口车辆检测实验结果表明，该方法在机动车与非机动车混行，公交车与小汽车相互遮挡的交叉口场景下，能较准确地得到车辆检测结果。 

A Vehicle Detecting Method Based on Pattern Recognition Combined with ST-MRF

The main difficulty of the vehicle detecting technology lies in overcoming the inaccuracy of detecting, which is directly affected by video shaking, mutual occlusion between vehicles, occlusion between the vehicles, and its shadow due to illumination changes. Based on the traditional spatial-temporal Markov random field (ST-MRF), the vehicle detecting method combined with pattern recognition and ST-MRF is proposed to address the problem. The boundary between two vehicles occluding each other is segmented by the pattern recognition technology, with the edge clearances and boundary information of occluded vehicles identified. Then, the results of pattern recognition are fed back to ST-MRF algorithm, which reassigns labels to occluded vehicles, integrates incomplete segmentation, and determines individual vehicle information. The results in the road section show 325 vehicles driving in the test area, vehicles tracked by the original ST-MRF algorithm, and the success rate is 79%. When the pattern recognition technology combined with ST-MRF algorithm to calculate 315 vehicles, the success rate is 97%. The results at the intersection show that the method can obtain more accurate results of vehicle detection at the intersection where motor vehicles and non-motor vehicles are mixed, and buses and cars are obstructed by each other.