我正在使用Rekognition的StartPersonTracking和GetPersonTracking与某些人进行视频分析。结果是一个json文件,如下所示:
"Persons": [{
"Timestamp": 0,
"Person": {
"BoundingBox": {
"Width": 0.0520833320915699,
"Top": 0.3125,
"Left": 0.40625,
"Height": 0.09444444626569748
},
"Index": 4
}
},
{
"Timestamp": 33,
"Person": {
"BoundingBox": {
"Width": 0.20624999701976776,
"Top": 0.4611110985279083,
"Left": 0.17499999701976776,
"Height": 0.3986110985279083
},
"Index": 1
}
},
{
"Timestamp": 33,
"Person": {
"BoundingBox": {
"Width": 0.22291666269302368,
"Top": 0.3194444477558136,
"Left": 0.7708333134651184,
"Height": 0.5444444417953491
},
"Index": 2
}
},
{
"Timestamp": 33,
"Person": {
"BoundingBox": {
"Width": 0.12083332985639572,
"Top": 0.36666667461395264,
"Left": 0.4520833194255829,
"Height": 0.24722221493721008
},
"Index": 3
}
},
{
"Timestamp": 33,
"Person": {
"BoundingBox": {
"Width": 0.0520833320915699,
"Top": 0.3125,
"Left": 0.40625,
"Height": 0.09444444626569748
},
"Index": 4
}
},
{
"Timestamp": 100,
"Person": {
"BoundingBox": {
"Width": 0.24375000596046448,
"Top": 0.2888889014720917,
"Left": 0.5833333134651184,
"Height": 0.5847222208976746
},
"Index": 0
}
},...
每个项目都有其时间戳,因此我们可以在整个视频中跟踪每个人。问题在于两次检测之间的差距可能很大。有没有已知的减小间隙的方法,即提高检测密度?
我在文档中或在php / java SDK中都找不到任何内容