Table 2 Computer vision-based positioning, localizing and scene recognizing systems

Huang et al. [62]

Indoor positioning

3D signature of places for feature detection, Novel K-locations algorithm

(+) 90% of the exposed errors are within 25 cm and 2° for location and orientation respectively

Kawaji et al. [65]

Indoor positioning

PCA-SIFT features and locality sensitive hashing

(+) Running time reduced while comparing with the pure SIFT features-based system

Deniz et al. [67]

Localization using texts in boards and banners

Canny edge detector, Tesseract and ABBY fine reader OCRs

ABBY fine reader showed better recognition rate than Tesseract

Adorno et al. [78]

Floor detection method

Superpixel segmentation and Hough line transform

(+) Accuracy: 87.6% for the unstructured environment and 93% for the structured environment

Murillo et al. [61]

Personal localization in indoor areas

GIST and SURF-based feature detector, Extended Kalman filter monocular SLAM

(+) 82% correct localization

Xiao et al. [68]

Indoor positioning in large indoor areas

CNN and SIFT features

(+) Low cost, accuracy: less than 1 m

Chen et al. [71]

Indoor positioning

CNN and ORB features

(+) Average position error: less than 0.35 m

Kendall et al. [75]

Indoor and outdoor localization

CNN

(+) Robust to various lighting and motion blur scenarios

Bashiri et al. [79]

Indoor object recognition to assist people with VI

Transfer learning based on the CNN model (AlexNet)

(+) Accuracy: 98%

Jayakanth [81]

Indoor object recognition to assist people with VI

CNN and texture features

(+) Accuracy: 100%

Afif et al. [82]

Indoor object detection to assist people with VI

CNN

Mean average precision: 84.16%