区域入侵检测

指定入侵区域，输入视频流，检测视频中人体位置，对人体进行跟踪，判断人体是否进入或离开入侵区域，如进入或来开入侵区域则报警，人体进区域后在区域内活动不重复报警。
精度对应场景或者限制条件
1）视频分辨率：不低于720P，码率≥1Mbps，建议2Mbps
2）环境要求：室内/室外，画面中光线均匀，无强背光、无强逆光，无过亮过暗现象，目标清晰人眼可见。
3）算法边界：
识别尺寸：最小人体像素>64*32(1080P)，64为长边32为短边，人体至少1/2未被遮挡，检测目标清晰人眼可见;
尺寸：以1920*1080分辨率图像为基准，人体目标最小像素为长短边64*32px；

Specify the intrusion area, input the video stream, detect the positions of human bodies in the video, track the human targets, determine whether a human enters or leaves the intrusion area, and trigger an alarm when a human enters or leaves the area. No repeated alarms will be activated if a human stays and moves within the area after entering it.

Scenarios and constraints for accuracy performance:
1. Video Resolution: No less than 720P, bitrate ≥ 1 Mbps, 2 Mbps is recommended.
2. Environmental Requirements: Applicable to indoor and outdoor scenarios. The video frame shall have uniform lighting, no strong backlight, no strong contre-jour, no overexposure or underexposure, and all targets shall be clearly visible to the naked eye.
3. Algorithm Boundaries:
- Recognition Size Requirement: For 1080P resolution, the minimum pixel size of a human body shall be >64×32, where 64 is the long side and 32 is the short side. At least 1/2 of the human body shall not be occluded, and the detection target shall be clearly visible to the naked eye.
- Size Reference Benchmark: Based on 1920×1080 resolution images, the minimum pixel size of a human target is 64×32px (long side × short side).

搜集汇总

背景与挑战

背景概述

该数据集是一个用于区域入侵检测的算法模型，能够通过视频流检测和跟踪人体位置，判断是否进入或离开指定区域并触发报警。数据集对视频分辨率、环境光线和目标清晰度有特定要求，适用于室内外监控场景。