停车场为了监管车辆停放，往往采用大屏实时视频展示人工观察的方式。但这种方法存在着工作量大和被动依赖工作人员等引起的监管不到位问题。为此，本文以人工智能技术在车辆停放管理中的应用为切入点，将神经网络的相关模型与监控技术相结合，研究并给出了一套能够按照划定区域识别不同类别的车辆违法驶入或停放在禁停区的智能识别方法；同时借助于所研究的方法，实现了一个能够应用于停车场辅助管理的车辆智能识别与停放监管软件。本文主要研究工作及内容概要如下:

(1) 针对YOLO网络识别微小目标困难、不适应远距离摄像头画面的缺点，论文给出了YOLOv5-AT车辆智能识别方法。将微小目标检测头与注意力模块融入YOLOv5神经网络结构，在神经网络的Neck部分融合Bi-FPN结构，增强网络的特征融合能力，提高网络识别的准确度，使神经网络更加适应摄像头拍摄的画面。通过网络性能对比实验，证明YOLOv5-AT网络具有良好的性能，在准确性上优于传统YOLO网络结构。

(2) 针对传统动态目标监测算法与神经网络结合后速度缓慢的问题，论文提出DIOU动态目标监测算法。该算法利用神经网络预测框进行动态目标过滤，速度快，准确度高，防止无效信息充斥于软件画面，达到了使神经网络在停车场环境良好运用的目的。通过实验证明，该算法在结合神经网络使用的前提下速度与准确性均优于传统动态目标监测算法。

(3) 设计并实现了一套车辆停放智能监管软件。用户通过使用该软件可以维护“禁行区域”，并选择禁行区域的类型。用户划定的禁行区域经过坐标换算后会被传递给车辆位置监管程序，程序判断会依照坐标判断是否有车辆进入禁行区域。当与选定类型相同的车辆驶入禁行区域时，软件中会记录违法车辆信息并给出提示。利用静态目标过滤算法，车辆只有在较快速度移动或者驶入禁止区域时才会被在画面上标注，降低了误判误报的风险。此外，用户还能结合实际情况调整静态目标过滤的强度，以适应各种不同的情况。

In the parking-lot, to supervise the event of illegal parking of vehicles often use the large screen real-time video display manual observation, but this method has a large workload and passive reliance on staff, easy to cause the problem of inadequate supervision. To this end, this thesis takes the application of artificial intelligence technology in vehicle parking management as the entry point, combines the neural network model with the monitoring technology, researches, and gives a set of intelligent recognition methods that can identify different categories of vehicles illegally driving or parking in the no-parking area according to the designated area; at the same time, with the help of the researched method, realizes a vehicle intelligent recognition and parking that can be applied to the parking lot auxiliary management. The main research work and content of this thesis are summarized as follows:

I. The thesis gives the YOLOv5-AT vehicle intelligence recognition method for the disadvantages of YOLO network which is difficult to recognize tiny targets and does not adapt to long-distance camera images. The tiny target detection head and attention module are integrated into the YOLOv5 neural network structure, and the Bi-FPN structure is fused in the Neck part of the neural network to enhance the feature fusion capability of the network, improve the accuracy of the network recognition, make the neural network more adaptable to the images taken by the camera, and allow the network to learn faster and converge faster. Through network performance comparison experiments, it is proved that the YOLOv5-AT network has good performance and outperforms the traditional YOLO network structure in terms of accuracy.

II. For the problem of the slow speed of traditional dynamic target monitoring algorithms combined with a neural network, the thesis proposes a DIOU dynamic target monitoring algorithm. The algorithm utilizes neural network prediction bounding boxes for dynamic target filtering, which is fast and accurate and prevents invalid information from flooding the system screen, achieving the goal of making good use of neural networks in the parking lot environment. Through experiments, it is proved that the algorithm is faster and more accurate than the traditional dynamic target monitoring algorithm when used in combination with neural networks.

III. Designed and implemented a set of vehicle parking intelligent supervision software. By using this software, users can maintain a "no-go zone" and select the type of no-go zones. The user-defined no-go area will be passed to the vehicle location monitoring program after coordinate conversion, and the program will judge whether a vehicle enters the no-go area according to the coordinates. When a vehicle of the same type as the selected one drives into the no-go zone, the information of the offending vehicle is recorded in the system and an alert is given. Due to the presence of a static target filtering algorithm in the system, the vehicle will only be marked on the screen when it moves at a faster speed or drives into a prohibited area, reducing the risk of misjudgment and false alarm. In addition, users can adjust the intensity of static target filtering in conjunction with the actual situation to adapt to various different road traffic conditions.

[1] 商讯.2019上半年全国机动车保有量达3.4亿辆[J].商用汽车,2019(07):7.

[2] 专家预计2035年中国新能源车保有量超1.6亿辆[J].重型汽车,2020(05):1.

[3] Computer Vision–ACCV 2020: 15th Asian Conference on Computer Vision, Kyoto, Japan, November 30–December 4, 2020, Revised Selected Papers, Part V[M]. Springer Nature, 2021.

[4] AI and machine learning paradigms for health monitoring system: intelligent data analytics[M]. Springer, 2021.

[5] Felici-Castell S, García-Pineda M, Segura-Garcia J, et al. Adaptive live video streaming on low-cost wireless multihop networks for road traffic surveillance in smart cities[J]. Future Generation Computer Systems, 2021, 115: 741-755.

[6] Sagaya Selvaraj A, Anusha S. RFID Enabled Smart Data Analysis in a Smart Warehouse Monitoring System using IoT[C]//Journal of Physics Conference Series. 2021, 1717(1): 012022.

[7] 邝豪, 高小岩. 智能化语音监控系统在群堆电厂的应用[C]. 生态互联数字电力——2019 电力行业信息化年会论文集, 2019.

[8] 朱迎春. 基于智能视频分析的视频监控系统研究[D]. 广州: 中山大学, 2010.

[9] 刘硕, 吴兆跃, 袁家政, 等. 物体定位与识别研究综述[C]. 中国计算机用户协会网络应用分会 2017 年第二十一届网络新技术与应用年会论文集, 2017.

[10] 吴宁. 基于流媒体技术的视频监控与分析系统[D]. 湖南大学, 2007.

[11] 郭加贝. 基于Android智能手机视频监控的应用软件开发[D]. 杭州电子科技大学, 2011.

[12] 周从亚. 视频监控系统中媒体流传输技术研究与实现[D]. 南京邮电大学, 2010.

[13] Boukerche A, Siddiqui A J, Mammeri A. Automated vehicle detection and classification: Models, methods, and techniques[J]. ACM Computing Surveys (CSUR), 2017, 50(5): 1-39.

[14] Tan T N, Baker K D. Efficient image gradient based vehicle localization[J]. IEEE Transactions on Image Processing, 2000, 9(8): 1343-1356.

[15] Burnos P. Alternative automatic vehicle classification method[J]. Metrology and Measurement Systems, 2010 (3).

[16] Arróspide J, Salgado L, Camplani M. Image-based on-road vehicle detection using cost-effective histograms of oriented gradients[J]. Journal of Visual Communication and Image Representation, 2013, 24(7): 1182-1190.

[17] Zhang H, Pan Z. Cross-voting SVM method for multiple vehicle classification in wireless sensor networks[J]. Sensors, 2018, 18(9): 3108.

[18] Mithun N C, Rashid N U, Rahman S M M. Detection and classification of vehicles from video using multiple time-spatial images[J]. IEEE Transactions on Intelligent Transportation Systems, 2012, 13(3): 1215-1225.

[19] Wang W. A Study on Contour Feature Algorithm for Vehicle Type Recognition[C]. International Joint Conference on Artificial Intelligence, 2009:452-455.

[20] Tao J, Chen S, Yang L, et al. Ultrasonic technique based on neural networks in vehicle modulation recognition[C]//Proceedings. The 7th International IEEE Conference on Intelligent Transportation Systems (IEEE Cat. No. 04TH8749). IEEE, 2004: 201-204.

[21] Zhan W, Luo Z. System design of real time vehicle type recognition based on video for windows (AVI) files[C]//International Conference on Intelligent Computing and Information Science. Springer, Berlin, Heidelberg, 2011: 681-686.

[22] Sarfraz M S, Saeed A, Khan M H, et al. Bayesian prior models for vehicle make and model recognition[C]//Proceedings of the 7th International Conference on Frontiers of Information Technology. 2009: 1-6.

[23] 李慧慧. 基于深度卷积神经网络的雾霾环境下车型识别算法研究[D]. 中北大学, 2017.

[24] 刘舒娴, 林伟, 陆培民. 基于卷积神经网络的车型识别研究[J]. 有线电视技术, 2017 (12): 88-91.

[25] 李新叶, 黄腾. 基于多尺度跃层卷积神经网络的精细车型识别[J]. 科学技术与工程, 2017, 17(11): 246-249.

[26] 王秀席, 王茂宁, 张建伟, 等. 基于改进的卷积神经网络 lenetG5 的车型识别方法[J]. 计算机应用研究, 2018, 35(7): 1G5.

[27] 林盛梅. 基于深度学习的车辆检测及车型分类方法研究[D]. 东南大学, 2018.

[28] 高思雯. 基于人工智能的大型载重汽车识别系统设计及实现[D]. 东北师范大学, 2019.

[29] Rosenblatt F. The perceptron: a probabilistic model for information storage and organization in the brain[J]. Psychological review, 1958, 65(6): 386.

[30] LeCun Y, Boser B, Denker J S, et al. Backpropagation applied to handwritten zip code recognition[J]. Neural computation, 1989, 1(4): 541-551.

[31] Cubuk E D, Zoph B, Mane D, et al. Autoaugment: Learning augmentation policies from data[J]. arXiv preprint arXiv:1805.09501, 2018.

[32] Ruder S. An overview of gradient descent optimization algorithms[J]. arXiv preprint arXiv:1609.04747, 2016.

[33] Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 779-788.

[34] Redmon J, Farhadi A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018.

[35] Bochkovskiy A, Wang C Y, Liao H Y M. Yolov4: Optimal speed and accuracy of object detection[J]. arXiv preprint arXiv:2004.10934, 2020.

[36] Tan M, Pang R, Le Q V. Efficientdet: Scalable and efficient object detection[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 10781-10790.

[37] Krause J, Stark M, Deng J, et al. 3d object representations for fine-grained categorization[C]//Proceedings of the IEEE international conference on computer vision workshops. 2013: 554-561.

[38] Tabassum S, Ullah S, Al-Nur N H, et al. Poribohon-BD: Bangladeshi local vehicle image dataset with annotation for classification[J]. Data in brief, 2020, 33.

[39] Hanji P, Alam M Z, Giuliani N, et al. HDR4CV: High Dynamic Range Dataset with Adversarial Illumination for Testing Computer Vision Methods[C]//London Imaging Meeting. Society for Imaging Science and Technology, 2021, 2021(1).