ConvNet and LSH-Based Visual Localization Using Localized Sequence Matching

. 2019 May 28;19(11):2439. doi: 10.3390/s19112439

Algorithm 1 ConvNet feature extraction and comparison.

Inputs:

{I_{i}^{t r a i n}}_{i = 1}^{N^{t r a i n}}

{training images database};

{I_{i}^{t e s t}}_{i = 1}^{N^{t e s t}}

{testing images database};

N^{t r a i n}, N^{t e s t}

{training and testing images numbers};
Outputs:
D {Cosine distance};
Algorithm:
for i ← 1 to

N^{t e s t}

do
for j ← 1 to

N^{t r a i n}

f_{j}^{t r a i n}

← Feature extraction for training images;

f_{i}^{t e s t}

← Feature extraction for testing images;

d_{i, j}

← cos 〈

f_{i}^{t e s t}

f_{j}^{t r a i n}

〉; // Cosine distance {Section 3.2}.
end for

D^{i}

←

[d_{i, 1}, d_{i, 2}, \dots, d_{i, N^{t r a i n}}]

; Column vector

D^{i} \in R^{N^{t r a i n} \times 1}

that contains the cosine distance between the testing image

I_{i}^{t e s t}

and all the training images {Section 3.2}.
end for