CN114928720A - System and method for detecting state of parking barrier gate rod - Google Patents

Abstract

The invention discloses a system and a method for detecting the state of a parking barrier gate rod, which comprises the following technologies and methods: in the use process of the barrier gate rod, more than 200 ten thousand pixel cameras are arranged in front of and behind the front (or back) of the barrier gate rod, and the angle and the position can be adjusted to completely cover the barrier gate rod in a video area; deploying image interception, an image preprocessing algorithm, an identification algorithm model and a road rod judgment method software to an edge computing terminal (located in a parking lot) or a cloud service; in the using process, an image is intercepted every second through an image intercepting algorithm, the area is calculated through an identification algorithm model, the state of the road rod is calculated through a road rod judging method, and early warning is carried out when the road rod is normally open, deformed and damaged. Through the mode, the system can help an unattended parking lot to find the running state and the damage condition of the barrier gate rod in time, improve the running efficiency of the parking lot and reduce the manual patrol and fee evasion conditions.

Description

Translated fromChinese

一种停车道闸杆的状态的检测系统及方法A system and method for detecting the state of a parking gate lever

技术领域technical field

本发明涉及智能停车场领域，特别是涉及一种道闸杆的状态的检测系统及方法。The invention relates to the field of intelligent parking lots, in particular to a system and method for detecting the state of a gate bar.

背景技术Background technique

无人停车方便了日常生活，提高了通行效率在日益发展的现代社会技术也愈发成熟；无人化是我们技术追求的标准、但一些潜在的问题还需要我们深入解决做到真正意义上的无人化。Unmanned parking facilitates daily life and improves traffic efficiency. In the increasingly developing modern society, technology is becoming more and more mature; unmanned is the standard pursued by our technology, but some potential problems still need to be solved in a real sense. Unmanned.

车牌系统已经具备较高无人智能化、但道闸这个在实际场景中很容易出现小车碰撞损坏的设备我们却无法感知它的运行状态。一般通常采用人工查看的方式但耗时费力效率低下，有方法利用在道闸安装传感器的方式来检测道闸状态但要适配不同道闸位置、安装要求高难度大，而且只能感知开启、关闭两种状态。也有方法[1]利用图像特征进行模板匹配，但存在以下几个缺点导致实用性很低：1.开启、关闭模板图像分别只有一帧，导致在白天晚上、雨雾晴天不同光照情况下特征发生较大变化从而模板匹配失效 2.未能精度定位道闸杆的精确像素坐标，从而提取了大量包含背景的特征干扰比对 3.只包含开启、关闭状态不能判断折断变形状态。The license plate system has a high level of unmanned intelligence, but the road gate, a device that is prone to collision and damage to cars in actual scenarios, we cannot perceive its operating status. Generally, manual viewing is usually used, but it is time-consuming, labor-intensive and inefficient. There are methods to detect the status of the barriers by installing sensors on the barriers, but it is difficult to adapt to different positions of the barriers, and the installation requirements are high and difficult, and they can only sense the opening and closing. Both states are closed. There is also a method [1] that uses image features for template matching, but there are several shortcomings that lead to low practicability: 1. There is only one frame for opening and closing the template image, which leads to the occurrence of features in different lighting conditions during the day and night, rain, fog and sunny days. Large changes make template matching fail. 2. The precise pixel coordinates of the gate bar cannot be accurately located, so a large number of feature interference comparisons including background are extracted. 3. Only the open and closed states can not judge the broken deformation state.

发明内容SUMMARY OF THE INVENTION

本发明主要解决的技术问题是提供一种道闸杆的状态的检测系统及方法，能够解决停车场无人值守道闸杆损坏或闸机故障等原因导致道闸杆不能及时关闭造成逃费问题。The main technical problem to be solved by the present invention is to provide a system and method for detecting the state of the gate bar, which can solve the problem of fee evasion caused by the failure of the gate bar to be closed in time due to the damage of the unattended gate bar in the parking lot or the failure of the gate machine. .

为解决上述技术问题，本发明采用的一个技术方案是：提供一种停车道闸杆的状态的检测系统及方法，包括以下技术和方法：在道闸杆使用过程中，在道闸杆正（或者斜）前（或后）面前后安装200万以上像素摄像头，调整角度和位置在视频区域里能完整覆盖到道闸杆；将图像截取、图像预处理算法、识别算法模型、道杆判断方法软件部署到边缘计算终端（位于停车场）或者云端服务；使用过程中，每秒钟通过图像截取算法截取图像，通过识别算法模型计算出面积，再通过道杆判断方法计算出道杆状态，遇到常开和变形及损坏进行预警。In order to solve the above-mentioned technical problems, a technical scheme adopted in the present invention is: to provide a detection system and method for the state of a parking gate lever, including the following technologies and methods: during the use of the gate lever, when the gate lever is ( Or oblique) front (or rear) front and rear cameras with more than 2 million pixels are installed, and the angle and position can be adjusted to completely cover the gate pole in the video area; image interception, image preprocessing algorithm, recognition algorithm model, and road pole judgment method The software is deployed to the edge computing terminal (located in the parking lot) or cloud service; during use, the image is intercepted by the image interception algorithm every second, the area is calculated by the recognition algorithm model, and then the road pole status is calculated by the road pole judgment method. Always open and early warning of deformation and damage.

本发明与现有技术相比，具有如下的优点和有益效果：Compared with the prior art, the present invention has the following advantages and beneficial effects:

现有技术主要通过人工对云端视频监控模式轮询查看道闸杆开关及变形及损坏情况，本发明能够通过人工智能算法自动实现预警无人值守停车场的道闸杆的运行状态和损毁情况，提高道闸杆正常状态的确认精确度，减少线上人工监控成本和线下人工巡逻成本及人员疲劳度、提高设备运行效率、减少逃费情况。The prior art mainly checks the switch, deformation and damage of the gate pole by manually polling the cloud video monitoring mode. The present invention can automatically realize early warning of the running state and damage of the gate pole of the unattended parking lot through artificial intelligence algorithms. Improve the accuracy of confirming the normal state of the gate pole, reduce the cost of online manual monitoring, offline manual patrol costs and personnel fatigue, improve equipment operation efficiency, and reduce toll evasion.

附图说明Description of drawings

图1是本发明一种道闸杆的状态的检测系统及方法的较佳实施例的立体结构示意图；1 is a schematic three-dimensional structural diagram of a preferred embodiment of a system and method for detecting the state of a gate lever of the present invention;

图2是一种道闸杆的状态的检测系统及方法所示的系统流程分解技术示意图；2 is a schematic diagram of a system flow decomposition technology shown in a system and method for detecting the state of a gate lever;

图3是一种道闸杆的状态的检测系统及方法所示的语义分割算法网络结构示意图；3 is a schematic diagram of the network structure of a semantic segmentation algorithm shown in a system and method for detecting the state of a gate pole;

图4是一种道闸杆的状态的检测系统及方法所示的基于ROI面积计算状态判断方法示意图。FIG. 4 is a schematic diagram of a state judgment method based on ROI area calculation shown in a system and method for detecting the state of a gate pole.

具体实施方式Detailed ways

下面结合附图对本发明的较佳实施例进行详细阐述，以使本发明的优点和特征能更易于被本领域技术人员理解，从而对本发明的保护范围做出更为清楚明确的界定。The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, and the protection scope of the present invention can be more clearly defined.

本发明实施包括：The implementation of the present invention includes:

1. 本发明提供基于图片识别的摄像头和道闸杆安装部署结构，在道闸杆正（或者斜）前（或后）面前后安装200万以上像素摄像头，调整角度和位置在视频区域里能完整覆盖到道闸杆，具体如图1较佳实施例的立体结构示意图所示。1. The present invention provides a camera and gate pole installation and deployment structure based on picture recognition, install more than 2 million pixel cameras in the front (or rear) front (or rear) of the gate pole, and adjust the angle and position in the video area. It completely covers the gate rod, as shown in the three-dimensional schematic diagram of the preferred embodiment in FIG. 1 .

2. 本发明提供道闸杆识别的系统解决方法处理流程如图2系统流程分解技术示意图所示，通过深度学习图像识别算法精准识别道闸杆状态用以解决减少线上人工监控成本和线下人工巡逻成本及人员疲劳度、提高设备运行效率、减少逃费情况。具体流程而言采用编解码模块抽取视频流的图像帧，然后利用滤波除杂、图像增强提升图像质量，再结合去雨去雾算法优化识别效果，接着利用图像语义分割算法精准识别道闸杆开启、关闭两种状态，同时提取图像中道闸杆的像素点集合，然后求取所有道闸杆像素点外接多边形坐标进而求出道闸杆面积，最后与预先采集标定的面积进行比对，如果当前面积小于标定面积则判断为破损变形。如果判断为变形异常情况则通知工作人员进行维护处理。2. The present invention provides a system solution method for gate pole identification. The processing flow is shown in Figure 2, the system flow decomposition technology schematic diagram. The deep learning image recognition algorithm is used to accurately identify the state of the gate pole to solve the problem of reducing the cost of online manual monitoring and offline monitoring. Manual patrol costs and personnel fatigue, improve equipment operation efficiency, and reduce evasion. Specifically, the codec module is used to extract the image frames of the video stream, and then filtering and image enhancement are used to improve the image quality, and then combined with the rain and fog removal algorithm to optimize the recognition effect, and then use the image semantic segmentation algorithm to accurately identify the opening of the gate. , close the two states, extract the pixel point set of the gate bar in the image at the same time, and then obtain the polygon coordinates of all the gate bar pixel points to obtain the gate bar area, and finally compare it with the pre-collected and calibrated area, if If the current area is less than the calibration area, it is judged as damaged and deformed. If it is judged that the deformation is abnormal, notify the staff to carry out maintenance processing.

3. 本发明提供语义分割网络结构如图3所示，使用ResNeXt-101残差网络作为特征提取网络,同时为了挖掘多尺度信息采用了FPN特征金字塔。ResNeXt-101和FPN的组合使得的特征挖掘表达能力更为强劲。骨干卷积网络Resnet-C4它将区域建议网络RPN生成的感兴趣区域ROI映射到Resnet-C4的输出中，并对其进行ROI Poling感兴趣区域池化操作，最后对池化后的结果进行并行分叉，分别对“class”类别、“box”边界框、“mask”分割掩码进行预测。3. The present invention provides a semantic segmentation network structure as shown in Figure 3, using ResNeXt-101 residual network as the feature extraction network, and at the same time using FPN feature pyramid in order to mine multi-scale information. The combination of ResNeXt-101 and FPN makes the feature mining expressive ability more powerful. The backbone convolutional network Resnet-C4 maps the region of interest ROI generated by the region proposal network RPN to the output of Resnet-C4, and performs the ROI Poling region of interest pooling operation on it, and finally parallelizes the pooled results. Fork, make predictions for the "class" category, the "box" bounding box, and the "mask" segmentation mask, respectively.

4. 本发明提供基于ROI面积计算道闸杆状态判断方法如图4所示，首先对输入的道闸杆视频流帧图片进行处理转化为网络需要的rgb格式，然后对全卷积语义分割网络模型进行初始化工作完成内存空间分配、网络参数加载操作，接着完成全尺寸的分割模型推理得到道闸杆的抬起或关闭状态、同时得到道闸杆不包含背景的ROI区域的具体像素点集，接下来对像素点集进行外接轮廓拟合获得多边形坐标，求取ROI多边形面积。我们预先已经标定计算好了道闸杆正常状态(开启、关闭)下的标定面积，由于正常状态无变形那么开启和关闭状态下的道闸杆像素点集合大小、外接轮廓面积是较为一致的。所以通过当前ROI面积与标定面积进行比对就能知道是否变形，不过在实际处理过程中道闸杆可能会有泥污、灰层等干扰，少量像素点预测不是那么精确所以我们引入阈值判断，也就是当前ROI面积与标定面积差的绝对值大于阈值时判断为变形，小于则为正常。4. The present invention provides a method for judging the state of the gate pole based on the calculation of the ROI area as shown in Figure 4. First, the input gate pole video stream frame picture is processed and converted into the rgb format required by the network, and then the fully convolutional semantic segmentation network is performed. The model is initialized to complete the memory space allocation and network parameter loading operations, and then complete the full-size segmentation model inference to obtain the lifted or closed state of the gate bar, and at the same time obtain the specific pixel set of the ROI area where the gate bar does not contain the background, Next, perform external contour fitting on the pixel point set to obtain the polygon coordinates, and obtain the ROI polygon area. We have calibrated and calculated the calibration area of the gate bar in the normal state (open, closed) in advance. Since there is no deformation in the normal state, the size of the pixel point set and the outer contour area of the gate bar in the open and closed states are relatively consistent. Therefore, by comparing the current ROI area with the calibration area, you can know whether it is deformed. However, in the actual processing process, the gate rod may have interference such as mud and gray layers. The prediction of a small number of pixel points is not so accurate, so we introduce a threshold value judgment. That is, when the absolute value of the difference between the current ROI area and the calibration area is greater than the threshold, it is judged as deformed, and when it is less than the normal value.

以上所述仅为本发明的实施例，并非因此限制本发明的专利范围，凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。The above descriptions are only the embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied to other related technologies Fields are similarly included in the scope of patent protection of the present invention.

Claims (3)

Translated fromChinese

1.一种停车道闸杆的状态的检测系统及方法，其特征在于，包括以下技术和方法：1. the detection system and method of the state of a parking road brake lever, is characterized in that, comprises following technology and method:A、 安装方法：在道闸杆正（或者斜）前（或后）面前后安装200万以上像素摄像头，调整角度和位置在视频区域里能完整覆盖到道闸杆；A. Installation method: Install a camera with more than 2 million pixels in front (or rear) of the front (or rear) of the gate pole, and adjust the angle and position to completely cover the gate pole in the video area;B、 图像截取：通过视频流解码模块截取视频流里图像帧，用于后续图像道闸杆识别处理；B. Image interception: intercept the image frame in the video stream through the video stream decoding module, which is used for the subsequent image gate identification processing;C、 图像预处理：为了提升识别效果我们一方面采用了滤波算法抑制噪点模糊，同时采用图像增强、去雨去雾算法解决极端天气条件下对识别带来的影响；C. Image preprocessing: In order to improve the recognition effect, we use a filtering algorithm to suppress noise and blur, and at the same time use image enhancement, rain and fog removal algorithms to solve the impact on recognition under extreme weather conditions;D、 识别算法模型：通过视频流截取道杆图像，通过全卷积深度网络进行像素级语义分割训练，训练出适合道闸杆（白天、晚上）图像分割算法模型，计算出道杆的几种状态（抬起、关闭）道杆的几何面积；D. Recognition algorithm model: intercept road pole images through video stream, perform pixel-level semantic segmentation training through fully convolutional deep network, train an image segmentation algorithm model suitable for road poles (day and night), and calculate several states of road poles (raised, closed) the geometric area of the pole;E、 道杆判断方法：通过与保存正确的道杆状态的面积进行比较，输出道杆状态（抬起、关闭、变形）；E. Judgment method of road pole: output the road pole state (lift, close, deform) by comparing with the area where the correct road pole state is saved;F、 在道闸杆使用过程：将图像截取、图像预处理、算法识别算法模型、道杆判断方法软件部署到边缘计算终端或者云端服务， 使用过程中，每秒钟通过图像截取算法截取图像，通过识别算法模型计算出面积，再通过道杆判断方法计算出道杆状态，遇到常开和变形及损坏进行预警。F. In the process of using the gate pole: Deploy the image capture, image preprocessing, algorithm recognition algorithm model, and road pole judgment method software to the edge computing terminal or cloud service. During the use process, the image is captured by the image capture algorithm every second. The area is calculated through the identification algorithm model, and then the pole status is calculated through the pole judgment method, and early warning is given when encountering normally open, deformation and damage.2.根据权利要求1所述的一种停车道闸杆的状态的检测系统及方法，其特征在于：图像截取方法，对于天气情况、照度等复杂环境下进行图像增强，使图像清晰度增加，同时利用去雨去雾算法解决极端气候场景识别问题。2. the detection system and method of the state of a kind of parking road gate rod according to claim 1, it is characterized in that: image interception method, carries out image enhancement under complex environment such as weather condition, illuminance, makes image sharpness increase, At the same time, the rain and fog removal algorithm is used to solve the problem of extreme climate scene recognition.3.根据权利要求1所述的一种停车道闸杆的状态的检测系统及方法，其特征在于：识别算法模型，通过全卷积深度网络学习后可以解决恶劣天气情况、白天晚上照度不同等复杂环境下的图像品质不同，图像分割算法模型得到优化，从而得到准确的面积。3. the detection system and method of the state of a kind of parking road gate rod according to claim 1, it is characterized in that: identification algorithm model, can solve bad weather condition, different illuminance in day and night etc. after full convolution deep network learning The image quality in complex environments is different, and the image segmentation algorithm model is optimized to obtain accurate areas.

Images