基于红外成像与YOLOv8的高速公路动物迁徙监测统计系统研究

Highway Animal Migration Monitoring and Statistical System Based on Infrared Imaging and YOLOv8

摘要: 针对我国部分边远地区高速公路因中大型动物迁徙引发的交通安全问题，提出并实现了一种基于可切换红外成像与深度学习的目标检测与统计系统。系统以立创·庐山派K230开发板为核心处理单元，搭载IR CUT红外成像摄像头进行信息采集。利用YOLOv8深度学习算法构建实时目标检测模型，对采集到的红外图像进行实时分析与推理，识别结果通过串口传输至上位机，实时显示道路周边野生动物活动情况并记录于日志文件中。实验模型评估中藏羚羊的AP (Average Precision)为0.961，表现优秀。野牦牛的AP为0.453，表现较差。野牦牛的正确识别率分别为95%和54%。实验结果表明本系统对动物进行识别统计的可行性，具备良好的实际应用潜力。

Abstract: In response to traffic safety issues caused by medium and large animal migrations on highways in some remote areas of China, a target detection and statistics system based on switchable infrared imaging and deep learning is proposed and implemented. The system uses the Lichuang·Lushanpai K230 development board as the core processing unit, equipped with an IR-CUT infrared imaging camera for information acquisition. Leveraging the YOLOv8 deep learning algorithm, a real-time object detection model is constructed to analyze and process the captured infrared images in real-time. The recognition results are transmitted to an upper computer via a serial port, enabling real-time display of wildlife activity around the road and recording in log files. In the model evaluation, the Tibetan antelope achieved an AP (Average Precision) of 0.961, indicating excellent performance. In contrast, the wild yak achieved an AP of 0.453, indicating relatively poorer performance. The correct recognition rates for the wild yak were 95% and 54%, respectively. Experimental results demonstrate the feasibility of the system for animal identification and statistics, showing good potential for practical application.