init_deviceid: "ALL" #例: 0; 0,1; 2,3; ALL # 基础控制参数 base: # 股道名称 track_name: "1" # 测试模式 test_model: false # Api 监听端口 api_port: 7070 # 是否上传识别结果 up_result: false # 是否启用socket-server use_socket_server: false # 日志文件目录 log_path: "./logs" # 识别结果目录 result_path: "./result" # 调试结果目录 debug_result_path: "./debug_result" # 最优识别目录 best_result_path: "./best_result" # 结果存储天数 result_save_days: 2 # 日志参数 log: # 输出日志级别[DEBUG, INFO, WARN, ERROR, FATAL] out_level: "DEBUG" # 保存日志级别 save_level: "DEBUG" #识别数据来源参数配置 data_source: #url: "rtsp://admin:sgt12345@10.27.119.13:554/h264/ch1/main/av_stream" url: "./vedio/buertai2.mp4" # 跳帧数 skip_interval: 3 # 识别目标 target: "NUM" # 行驶方向 0-自动识别 1-向左 2-向右 (与“首位信息”成对存在,形成例如向左就编号在前,向右就属性在前的对应) direction: 0 # 0-向左编号在前 1-向左属性在前 (向右行驶的情况:2-向右编号在前 3-向右属性在前) left_first: 0 # (向左行驶的情况:0-向左编号在前 1-向左属性在前) 2-向右编号在前 3-向右属性在前 right_first: 3 # (ltx,lty,rbx,rby) identify_areas: "120, 0, 1800, 1080" # 大框的最小高度(为屏蔽远股道识别到的信息) classid_minheight: "1:90, 2:120, 3:120, 9:240, 10:240, 18:120" # 识别参数 identify: # 硬件解码 hardware_decode: true # 运行方式 run_mode: "always" #[always; command] # 是否开启动态检测 need_move_detect_flag: true # 识别方向 [LEFT,RIGHT,ALL] identify_direction: "LEFT" # 大框帧跨度(比一个大框从出现到消失的跨度稍大一点, 跟跳帧有关系) partition_frame_span: 20 # 大框帧跨度的位置像素差异 split_frame_span_px: 200 # 每帧大框位置差异最小值 (持续小于此值,则可能停车) chkstop_px: 15 # 持续X次续位置差异小于chkstop_px,则判断为停车。 chkstop_count: 10 # 过滤最小大框高度(不需要的话就写个很小的值) num_frame_height: 150 pro_frame_height: 120 # 过滤最大框宽度(不需要的话就写个很大的值) space_frame_width: 500 # 是否识别车头 train_heard_detect: true # 选优 0-频率优先 1-长度优先 select_best_mode: 0 # 保存图片质量(1~100 越高越清晰) save_pic_quality: 50 #是否实时推流-用于直播 #gc_push_actual_flag: false # 模型参数 model: # 来车检测 MoveEngine: path: "./model/step0/step0.engine" model_info_path: "./model/step0/move_modelinfo.txt" score_threshold: 0.9 nms_threshold: 0.3 # 关键区域识别 TrainStepOneEngine: path: "./model/step1/step1.engine" model_info_path: "./model/step1/train_step1_modelinfo.txt" score_threshold: 0.6 nms_threshold: 0.3 # 字符识别 TrainStepTwoEngine: path: "./model/step2/step2.engine" model_info_path: "./model/step2/train_step2_modelinfo.txt" score_threshold: 0.7 nms_threshold: 0.3 # 定检期关键区域识别 ChkDateStepOneEngine: path: "./model/chkDate_step1/step1.engine" model_info_path: "./model/chkDate_step1/chkdate_step1_modelinfo.txt" score_threshold: 0.6 nms_threshold: 0.3 # 定检期字符识别 ChkDateStepTwoEngine: path: "./model/chkDate_step2/step2.engine" model_info_path: "./model/chkDate_step2/chkdate_step2_modelinfo.txt" score_threshold: 0.7 nms_threshold: 0.3 # 集装箱关键区域识别 StepOneContainerEngine: path: "./model/container_step1/con1.engine" model_info_path: "./model/container_step1/container_step1_modelinfo.txt" score_threshold: 0.6 nms_threshold: 0.3 # 集装箱字符识别 StepTwoContainerEngine: path: "./model/container_step2/con2.engine" model_info_path: "./model/container_step2/container_step2_modelinfo.txt" score_threshold: 0.7 nms_threshold: 0.3 # http 接口 http_client: # 服务器IP http_ip: 192.168.2.108 # 通讯端口 http_port: 20004 # 获取接口授权地址 token_path: "/api/blade-auth/oauth/token" # 识别结果上传地址 up_result_path: "/api/train-carriage/identification/rfid-save" # 设备状态上传地址 device_status_url: "/api/blade-train/deviceInfo/save" # 接口用户名 username: "guest_01" # 接口密码 password: "d55b0f642e817eea24725d2f2a31dd08" # 上传图片的地址 image_srv: "http://192.168.0.121:9010/" # websocket_server 的服务端参数 wsocket_server: port: 7071 max_queue_len: 10 #sftp用户名、密码、地址 ftp: type: "ftp" #可选 ftp 或 sftp ip: "192.168.2.138" port: 21 # ftp默认21 sftp默认22 username: "nvidia" password: "nvidia" image_path: "" quit_time: 10 #无上传任务延迟XXX秒断开FTP连接 gc_space_minrbx_imgpercent: 0 #间隔框最低点不应小于画面某个高度值(该值为画面百分比) [主要为屏蔽远股道间隔框,若不需要屏蔽则配置为0]