Object Detection Using Image Processing For Blind Persons: A Review

Abstract

The fundamental idea of detection of objects for videos is confirming an object\'s existence in a series of images and maybe pinpointing it specifically for identification. Monitoring an item\'s existence, location, size, form, and other physical and Changes in time throughout a video sequence is known as object tracking. The challenge of reproducing the designated area in consecutive frames of a set of photographs captured at closely spaced time intervals is referred to as the temporal correspondence problem. and it must be solved in order to do this. The two processes in question are interrelated. The foundation of tracking is detection, which often begins with the detection of objects. To aid and validate tracking, it is frequently required to identify an item repeatedly in a following image sequence. In light of this, the present review describes how object recognition using image processing might assist the blind.

Introduction

Object detection is a key area in computer vision, artificial intelligence, and deep learning, serving as a foundation for advanced tasks like event detection, behavior analysis, video surveillance, autonomous driving, medical imaging, and industrial inspection. The goal is to identify object categories, locate them within images, and define their bounding boxes.

1. Traditional vs. Deep Learning Approaches

• Traditional methods rely on manual steps: preprocessing, window sliding, feature extraction and classification, but suffer from issues like low accuracy, inefficiency, and sensitivity to changes.

• Deep learning, particularly using CNNs, has revolutionized object detection. The breakthrough came with AlexNet (2012), which outperformed traditional models on large datasets like ImageNet.

2. Evolution of CNN-based Detection Models

• R-CNN (2014): Introduced region proposals and used SVMs, but was slow and computationally heavy.

• SPP-net: Improved efficiency by extracting features once and pooling spatially, but retained R-CNN's multi-stage complexity.

• Faster R-CNN: Replaced region proposals with a Region Proposal Network (RPN), sharing convolutional layers with the detector for better speed and accuracy, but still not real-time.

3. One-Stage Detection Algorithms

• YOLOv1 (2016): Treated detection as a single regression problem, enabling real-time speed (~45 FPS), but had limitations in small object detection.

• YOLOv3: Improved small object detection using Darknet53 and multi-scale predictions, converting to multi-label classification.

• YOLOv4 (2020): Introduced by Alexey Bochkovskiy, combines advanced techniques (e.g., SPP, PANet, CSPDarknet53, Mish activation) for superior accuracy and speed balance.

4. Object Detection via Image Processing

• UAVs (Unmanned Aerial Vehicles): Widely used in defense and surveillance, require efficient and accurate object detection for tasks like ISR, combat rescue, and EW.

• UAVs traditionally use costly laser or multisensor systems. This research promotes image processing-based UAV detection to lower costs and improve efficiency.

5. Computer Vision Applications and Methodology

• Computer vision algorithms enable object recognition, categorization, and tracking in images/videos for applications such as robotics, photo organization, and medical imaging.

• Object detection locates and classifies objects using template matching and correlation techniques (e.g., SSD).

6. Image Processing and Face Detection

• Image processing involves converting images to digital form and analyzing them for features using 3 main steps: image input, modification/analysis, and output.

• Face detection uses techniques like the Viola-Jones algorithm with cascade classifiers for tracking facial features, though accuracy can drop with head tilts or occlusions.

7. Research Scope

• The project aims to improve object detection on UAVs using affordable image processing methods instead of costly sensors, ensuring reliability, reducing human workload, and compensating for detection errors.

Conclusion

This post was created from the ground up, covering everything from the necessity of computer vision to the methods and rationale behind object and face detection. We went into great depth on every idea related to object and face detection, as well as the significance of this subject. Our findings demonstrated that the primary goal was object detection, and the output items were identified from the actual picture. In surveillance and other contexts, the face detection software can be used to identify and track individuals. To help understand the effectiveness of image detecting structures, we presented a tool. We introduce strategies for visualizing object detector success spaces. The findings of Paul Viola and Michael Jones (2001) corroborate ours. In order to validate findings, we also presented and assessed a number of techniques for visualizing object identification attributes.

References

[1] Wu, R.B. Study on Face Recognition and Intelligent Video Surveillance Applications in Prison Security. 2019; 6: 16–19; China Security Technology and Application. [2] Research and Challenge of Deep Learning Methods for Medical Image Analysis by Tian, J.X., Liu, G.C., Gu, S.S., Ju, Z.J., Liu, J.G., Gu, D.D. In 2018, ActaAutomaticaSinica 44: 401-424. [3] The research status and development trend of intelligent detection technology and industrial robot target identification are presented by Jiang, S.Z., and Bai, X. 2020; 36: 65– 66 Guangxi Journal of Light Industry. [4] Deep Convolutional Neural Networks for ImageNet Classification by A. Krizhevsky, I. Sutskever, and G. Hinton. 2012; 25: 1097–1105, Advances in Neural Information Processing Systems. [5] ImageNet Large Scale Visual Recognition Challenge / Russakovsky, O., Deng, J., Su, H., et al. 2015; 115: 211-252 in International Journal of Computer Vision. [6] Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation Girshick, R., Donahue, J., Darrel, T., Malik, J. In: Columbus, 2014, Computer Vision and Pattern Recognition, pp. 580–587. [7] Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition by He, K.M., Zhang, X.Y., Ren, S.Q., and Sun, J. 2015; 37: 1904–1916; IEEE Transactions on Pattern Analysis & Machine Intelligence. [8] Fast R-CNN by R. Girshick is published in the IEEE International Conference on Computer Vision Proceedings. 2015 Santiago, pp. 1440–1448. [9] Faster R-CNN: towards real-time object identification using region proposal networks, Ren, S.Q., He, K.M., Girshick, R., Sun, J. In: Neural Information Processing Systems Advances. Montre al, 2016; pages 91–99. [10] Farhadi, A., Grishick, R., Divvala, S., and Redmon, J. Unified, Real-Time Object Detection—You Only Need to Look Once. In: Las Vegas, 2016; Computer Vision and Pattern Recognition, pp. 779–788. [11] Redmon, J. and Farhadi, A. Yolo9000: enhanced, accelerated, and fortified. In: Hawaii.2017, Computer Vision and Pattern Recognition, pp. 7263-7271. [12] Farhadi, A. and Redmon, J. (2018) Yolov3: A step at a time. ArXiv: Pattern Recognition and Computer Vision. [13] SSD: Single Shot MultiBox Detector Liu, W., Anguelov, D., Erhan, D., et al. 2016, pp. 2137, European Conference on Computer Vision. [14] YOLOv4: Ideal Speed and Accuracy of Object Detection, Bochkovskiy, A., Wang, C.Y., Liao, H.Y.M. 2020 ArXiv: Pattern Recognition and Computer Vision [15] [Online] Available: scholar.google.fr/scholar?hl = fr&q = Object + detection + using + Haar cascade + Classifier&btnG = &lr = [16] [Online] Available: lab.cntl.kyutech.ac.jp/ kobalab/ nishida/opencv/OpenCVObjectDetectionHowT o.pdf [17] [Online] Available: cs.colby.edu/maxwell/courses/cs397 vision/F 07/papers/viola ? Faces ? cvpr01.pdf [18] [Online] Available: cbcl.mit.edu/publications/ps/heisele x3hei.lo.pdf [19] [Online] Available: arxiv.org/pdf/1502.05461v1.pdf [20] [Online]Available:www.researchgate.net/profile/VitorSantos6/publication/267868 2 AutomaticDetectionof Cars in Real Roads using Haar?like Features/links/552c0bae0cf 2e089a3ac3bc3.pdf [21] [Online] Available:www.svcl.ucsd.edu/projects/traffic/ [22] [Online] Available: github.com/andrewssobral/bgslibrary#bgslibrary [23] [Online] Available: github.com/andrewssobral/simple vehicle counting [24] [Online] Available: github.com/andrewssobral/bgslibrary [25] C.P. Papageorgiou, M. Oren, T. Poggio, A general framework for object detection, in: ICCV ’98: Proceedings of the Interna- tional Conference on Computer Vision, Washington, DC, USA, 1998, pp. 555-562. [26] S.Z. Li, L. Zhu, Z. Zhang, A. Blake, H. Zhang, H. Shum, Statistical learning of multiview face detection, in: ECCV ’02: Proceedings of the European Conference on Computer Vision, Lecture Notes in Computer Sciences, vol. 2353, London, UK, 2002, pp. 67-81. [27] R. Lienhart, J. Maydt, An extended set of Haar-like features for rapid object detection, in: [28] ICIP ’02: Proceedings of the International Conference on Image Processing, 2002, pp. 900- 903 [29] P. Viola, M.J. Jones, Rapid object detection using a boosted cascade of simple features, in: CVPR ’01: Proceedings of the Conference on Computer Vision and Pattern Recognition, Los Alamitos, CA, USA, 2001, pp. 511-518. [30] M. Jones, P. Viola, Fast multi-view face detection, Technical Report MERLTR2003-96, Mitsubishi Electric Research Labora- tories, July 2003. [31] Viola, P.; Jones, M., ”Rapid object detection using a boosted cascade of simple features,” Computer Vision and Pattern Recog- nition (CVPR), 2001 IEEE Conference on, June 2001. [32] [Online]Available:ir.library.louisville.edu/cgi/viewcontent.cgi?a rticle = 2731&context = etd [33] “A Machine Learning Approached Model to Identify the Object for Visually Impaired Person(2023)”by Sunita Joshi ,Neha Gupta, Mitali, Gautam Yadav [34] John Melonakosa and Yi Gaoa and Allen Tannenbauma, Tissue Tracking: Applications for Brain MRI Classification , Georgia Institute of Technology,14 Ferst Dr, Atlanta, GA, USA [35] KARMA, Computer

Copyright

Copyright © 2025 Aditya Gandhi, Ayush , Devansh Srivastava, Ambuj Tripathi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.