本次提交实现了完整的人员行为分析系统,包括: 1. 新增基于位置和跟踪ID的两种行为检测算法 2. 新增徘徊检测服务与行为处理器模块 3. 前后端集成算法配置界面与告警展示 4. 支持图片和视频流场景下的行为分析 5. 新增算法配置接口与文档说明 具体改动: - 新增loitering_detection模型目录与算法实现 - 新增AlgorithmConfig组件实现可视化配置 - 扩展图片/视频检测接口支持算法参数传递 - 新增行为告警推送与前端展示页面 - 优化检测服务,集成行为分析逻辑 - 移除冗余日志输出,完善代码注释
237 lines
7.6 KiB
Python
237 lines
7.6 KiB
Python
"""
|
||
基于位置的静止人员检测算法
|
||
不依赖跟踪 ID,而是根据位置来关联人员
|
||
适用于跟踪不稳定但人员相对静止的场景
|
||
"""
|
||
|
||
import time
|
||
from typing import Dict, List, Tuple, Optional
|
||
from dataclasses import dataclass, field
|
||
from collections import defaultdict
|
||
|
||
|
||
@dataclass
|
||
class PositionRecord:
|
||
"""位置记录"""
|
||
first_seen: float
|
||
last_seen: float
|
||
center: Tuple[int, int]
|
||
box: Tuple[int, int, int, int]
|
||
duration: float = 0.0
|
||
|
||
|
||
class PositionBasedStationaryDetector:
|
||
"""
|
||
基于位置的静止检测器
|
||
|
||
特点:
|
||
- 不依赖跟踪 ID,直接用位置关联人员
|
||
- 适用于 SORT 等跟踪器不稳定的场景
|
||
- 使用网格化位置 + 距离容差进行匹配
|
||
"""
|
||
|
||
def __init__(
|
||
self,
|
||
stationary_threshold: float = 10.0, # 静止阈值(秒)
|
||
position_tolerance: int = 50, # 位置容差(像素)
|
||
cleanup_interval: float = 5.0 # 清理间隔(秒)
|
||
):
|
||
self.stationary_threshold = stationary_threshold
|
||
self.position_tolerance = position_tolerance
|
||
self.cleanup_interval = cleanup_interval
|
||
|
||
# 位置历史记录: {position_key: PositionRecord}
|
||
self._position_history: Dict[Tuple[int, int], PositionRecord] = {}
|
||
self._last_cleanup = time.time()
|
||
|
||
def _get_position_key(self, center: Tuple[int, int]) -> Tuple[int, int]:
|
||
"""
|
||
将连续坐标转换为离散的位置键
|
||
用于将相近位置归为一类
|
||
"""
|
||
x, y = center
|
||
grid_x = int(x / self.position_tolerance)
|
||
grid_y = int(y / self.position_tolerance)
|
||
return (grid_x, grid_y)
|
||
|
||
def _find_matching_position(
|
||
self,
|
||
center: Tuple[int, int]
|
||
) -> Optional[Tuple[int, int]]:
|
||
"""
|
||
查找与当前位置匹配的历史位置
|
||
返回匹配的位置键,如果没有则返回 None
|
||
"""
|
||
current_key = self._get_position_key(center)
|
||
|
||
# 首先检查精确匹配
|
||
if current_key in self._position_history:
|
||
hist_center = self._position_history[current_key].center
|
||
distance = ((center[0] - hist_center[0]) ** 2 +
|
||
(center[1] - hist_center[1]) ** 2) ** 0.5
|
||
if distance < self.position_tolerance:
|
||
return current_key
|
||
|
||
# 检查相邻网格
|
||
for dx in [-1, 0, 1]:
|
||
for dy in [-1, 0, 1]:
|
||
if dx == 0 and dy == 0:
|
||
continue
|
||
neighbor_key = (current_key[0] + dx, current_key[1] + dy)
|
||
if neighbor_key in self._position_history:
|
||
hist_center = self._position_history[neighbor_key].center
|
||
distance = ((center[0] - hist_center[0]) ** 2 +
|
||
(center[1] - hist_center[1]) ** 2) ** 0.5
|
||
if distance < self.position_tolerance:
|
||
return neighbor_key
|
||
|
||
return None
|
||
|
||
def update(
|
||
self,
|
||
center: Tuple[int, int],
|
||
box: Tuple[int, int, int, int]
|
||
) -> Tuple[str, float, bool]:
|
||
"""
|
||
更新位置信息
|
||
|
||
Args:
|
||
center: (x, y) 中心点坐标
|
||
box: (x1, y1, x2, y2) 边界框
|
||
|
||
Returns:
|
||
position_id: 位置 ID(用于关联)
|
||
stationary_duration: 静止时长(秒)
|
||
is_stationary: 是否静止超过阈值
|
||
"""
|
||
current_time = time.time()
|
||
|
||
# 定期清理旧记录
|
||
if current_time - self._last_cleanup > self.cleanup_interval:
|
||
self.cleanup_old_positions()
|
||
self._last_cleanup = current_time
|
||
|
||
# 查找匹配的历史位置
|
||
matching_key = self._find_matching_position(center)
|
||
|
||
if matching_key is not None:
|
||
# 更新已有位置
|
||
record = self._position_history[matching_key]
|
||
record.last_seen = current_time
|
||
|
||
# 平滑更新中心位置(使用移动平均)
|
||
old_center = record.center
|
||
record.center = (
|
||
int(0.7 * old_center[0] + 0.3 * center[0]),
|
||
int(0.7 * old_center[1] + 0.3 * center[1])
|
||
)
|
||
record.box = box
|
||
|
||
duration = current_time - record.first_seen
|
||
record.duration = duration
|
||
|
||
is_stationary = duration > self.stationary_threshold
|
||
position_id = f"pos_{matching_key[0]}_{matching_key[1]}"
|
||
|
||
return position_id, duration, is_stationary
|
||
else:
|
||
# 创建新位置记录
|
||
new_key = self._get_position_key(center)
|
||
self._position_history[new_key] = PositionRecord(
|
||
first_seen=current_time,
|
||
last_seen=current_time,
|
||
center=center,
|
||
box=box,
|
||
duration=0.0
|
||
)
|
||
new_id = f"pos_{new_key[0]}_{new_key[1]}"
|
||
return new_id, 0.0, False
|
||
|
||
def cleanup_old_positions(self, max_age: float = 5.0) -> int:
|
||
"""
|
||
清理长时间未更新的位置记录
|
||
|
||
Args:
|
||
max_age: 最大保留时间(秒)
|
||
|
||
Returns:
|
||
清理的记录数量
|
||
"""
|
||
current_time = time.time()
|
||
to_remove = [
|
||
key for key, data in self._position_history.items()
|
||
if current_time - data.last_seen > max_age
|
||
]
|
||
|
||
for key in to_remove:
|
||
del self._position_history[key]
|
||
|
||
return len(to_remove)
|
||
|
||
def get_all_stationary(
|
||
self,
|
||
threshold: Optional[float] = None
|
||
) -> List[Dict]:
|
||
"""
|
||
获取所有静止超过阈值的位置
|
||
|
||
Args:
|
||
threshold: 静止阈值(秒),默认使用初始化时的阈值
|
||
|
||
Returns:
|
||
list: [{position_id, duration, center, box}, ...]
|
||
"""
|
||
threshold = threshold or self.stationary_threshold
|
||
|
||
result = []
|
||
for key, data in self._position_history.items():
|
||
if data.duration > threshold:
|
||
result.append({
|
||
'position_id': f"pos_{key[0]}_{key[1]}",
|
||
'duration': data.duration,
|
||
'center': data.center,
|
||
'box': data.box
|
||
})
|
||
|
||
# 按时长排序
|
||
result.sort(key=lambda x: x['duration'], reverse=True)
|
||
return result
|
||
|
||
def reset(self):
|
||
"""重置所有跟踪数据"""
|
||
self._position_history.clear()
|
||
self._last_cleanup = time.time()
|
||
|
||
def detect(
|
||
self,
|
||
detections: List[Dict]
|
||
) -> List[Dict]:
|
||
"""
|
||
批量检测静止状态
|
||
|
||
Args:
|
||
detections: 检测结果列表,每项包含 'bbox': [x1, y1, x2, y2]
|
||
|
||
Returns:
|
||
添加 'stationary_info' 字段的检测结果
|
||
"""
|
||
results = []
|
||
|
||
for det in detections:
|
||
x1, y1, x2, y2 = det['bbox']
|
||
center = ((x1 + x2) // 2, (y1 + y2) // 2)
|
||
box = (x1, y1, x2, y2)
|
||
|
||
position_id, duration, is_stationary = self.update(center, box)
|
||
|
||
det_copy = det.copy()
|
||
det_copy['stationary_info'] = {
|
||
'position_id': position_id,
|
||
'duration': round(duration, 2),
|
||
'is_stationary': is_stationary,
|
||
'threshold': self.stationary_threshold
|
||
}
|
||
results.append(det_copy)
|
||
|
||
return results
|