Algorithm 1: ATC Hand Gesture Recognition
Functions: ℱ = object detector; ℋ = hand detector; ℳ = 3D hand model; $\mathcal{E}$ = feature extractor; $\mathcal{C}$ = contrastive embedding; 𝔗 = single inference.
Input: videoframes (onboard device)
A = ℱ(Input)	// ATC detection
for each hand in f1 & f2:
H = ℋ(A)	// Hand detection
M = ℳ(H)	// 3D hand modeling
E = $\mathcal{E}$(M)	// Hand feature extraction
E = {Ef1, Ef2}	// Stores hand features (of Ef1, and Ef2)
$C$ = $\mathcal{C}$(E)	// Contrastive embedding calculation for Ef1, and Ef2
$I$ = 𝔗(Ef2)	// Inference for Ef2
$G=\{\begin{array}{c}C+I, if poses have difference\\ I, if poses have no difference\end{array}$
return ($G$)