施工场景下基于YOLOv3的安全帽佩戴状态检测

针对现有安全帽检测研究中采用的两阶段检测法存在检测效率偏低,累积误差对精度影响较大的问题,提出一种对安全帽的单阶段检测法.将安全帽和工人头部视为一个整体,将检测目标分为2类,即佩戴安全帽的头部和未佩戴安全帽的头部,同时对2类目标进行检测,避免了冗余的计算步骤及累积误差的影响.同时,针对施工场景安全帽佩戴状态检测特点,对...

Detection of wearing state of safety helmet based on YOLOv3 in construction scene

The two-stage detection method used in the existing safety helmet detection research has low detection efficiency, and the cumulative error has a greater impact on accuracy. In view of this problem, this paper proposed a single-stage detection method of safety helmets, which combined the safety helmet and the worker’s head. The department was regarded as a whole, and the detection targets were divided into two categories, heads with helmets and heads without helmets. The two types of targets were detected at the same time, avoiding the effects of redundant calculation steps and accumulated errors. At the same time, this paper improved the network structure, loss function and a prior boxes size of YOLOv3 based on the detection characteristics of the safety helmet wearing status in the construction scene, and the YOLOv3-C model was proposed based on the above. The experimental results show that the detection performance of YOLOv3-C model has been greatly improved. The m AP of the sample concentration model established in this paper reaches 93.84%.The average accuracy of the safety helmet detection reaches 97.01%.The average accuracy of the worker head detection reaches 90.67%. At the same time, YOLOv3-C shows good robustness to the detection scenario in this paper.