Algorithm 1 Entry and Departure Model Training Algorithm

Input: video V = <f1,f2,…,fn>, image classification algorithm set SUM_M, video interval time η, minimum tolerable noise rate ε. Output: Entry and departure recognition model Mt.

1: $R_i=\left\{i\right\}\times OD(f_i),\forall i=1,\dots ,n$, where $OD(f_i)$ indicates a lightweight object detection module 2: Define $S$ to store video image frame numbers and information after video image detection, $S={\displaystyle \underset{i=1}{\overset{n}{\cup }}{R}_i}$ 3: The video image frame number set $A=S.\mathrm{colomn}(1)$, where $\mathrm{colomn}(1)$ represents the extraction of the first component of the set 4: $\forall j\in A,{\alpha }_j=1\left(\exists k\in A(j-k<\eta \right.\left.)\right)$, 5: ${\phi }_j={\displaystyle \sum _{k=1}^{j-1}{\alpha }_j}$ 6: $\forall 1\le m\le n$, paragraph $m$ entry and departure training video $S_m=\left\{s_j\right|{\phi }_j=m\}$ 7: All generated tag collections SUM_TAG =${\displaystyle \underset{1\le m\le n}{\cup }\mathrm{Auto}-\mathrm{SD}(s_m)}$, where Auto-SD indicates the generated tag method; see Section 3.2 for details 8: All generated tag noise rates δ = Ad-EDB(SUM_TAG), where Ad-EDB denotes the tag conditioning evaluation method; see Section 3.3 for details 9: Image classification algorithm M = $\underset{size}{arc\min }\{M_i\in \mathrm{SUM}\_\mathrm{M}|\delta <\epsilon \}$ 10:  Entry and departure recognition model Mt = Train(M,S,SUM_TAG)