From classical techniques to convolution-based models: A review of object detection algorithms
1 st FNU Neha, 1 st Deepshikha Bhati, 2 nd Deepak Kumar Shukla, 3 rd Md Amiruzzaman · 2024
Read From classical techniques to convolution-based models: A review of object detection algorithms by 1 st FNU Neha, 1 st Deepshikha Bhati, 2 nd Deepak Kumar Shukla, 3 rd Md Amiruzzaman (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-05252.txt.
Introduction
Fnu Neha and co-authors review object detection methods in a 2024 paper. Object detection finds and labels items inside an image. Older methods used handcrafted features and shallow models.
Those methods struggled with complex scenes and missed high-level meaning. Deep learning, mainly convolutional networks, learns rich features from data itself. Features run from simple edges to full object concepts. The review traces the path from classical pipelines to modern convolution-based models. Strengths and limits of each era are compared. Readers get a structured map of a core computer vision task and how it matured.
Catalog record
| CREATOR | 1 st FNU Neha, 1 st Deepshikha Bhati, 2 nd Deepak Kumar Shukla, 3 rd Md Amiruzzaman |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-05252
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
From classical techniques to convolution-based models: A review of object detection algorithms Thanks: The first author contributed the most to this paper. Corresponding author: [email protected] 1 st FNU Neha, 1 st Deepshikha Bhati, 2 nd Deepak Kumar Shukla, 3 rd Md Amiruzzaman Abstract Object detection is a fundamental task in computer vision and image understanding, with the goal of identifying and localizing objects of interest within an image while assigning them corresponding class labels. Traditional methods, which relied on handcrafted features and shallow models, struggled with complex visual data and showed limited performance. These methods combined low-level features with contextual information and lacked the ability to capture high-level semantics. Deep learning, especially Convolutional Neural Networks (CNNs), addressed these limitations by automatically learning rich, hierarchical features directly from data. These features include both semantic and high-level representations essential for accurate object detection. This paper reviews object detection frameworks, starting with classical computer vision methods. We categorize object detection approaches into two groups: (1) classical computer vision techniques and (2) CNN-based detectors. We compare major CNN models, discussing their strengths and limitations. In conclusion, this review highlights the significant advancements in object detection through deep learning and identifies key areas for further research to improve performance. I INTRODUCTION Deep learning (DL) has advanced image analysis, especially in object classification, localization, and detection tasks. In classification, the aim is to assign an image or object within it to one of several categories [ 1 ] . However, classification does not provide the object’s location. Localization improves on this by identifying both the object’s category and position, typically with a bounding box [ 2 ] , though the precision of these boxes can vary. Object detection further extends classification and localization by detecting and classifying multiple objects in an image, providing bounding boxes for each [ 2 ] . The bounding box’s top-left corner is represented by (X_{min},Y_{min}) , and the bottom-right by (X_{max},Y_{max}) , along with a label indicating the object’s class as shown in Fig 1 . Fig. 1: (A) Single-object sunflower: A single bounding box localizes and classifies the central sunflower bloom. (B) Multiple-object sunflower: Multiple bounding boxes highlight and classify overlapping sunflowers and leaves, illustrating multi-scale object detection and localization within a complex scene. Object detection has applications across fields such as medical imaging, logo detection, facial recognition, pedestrian detection, and industrial automation. However, challenges arise from image transformations like changes in scale, orientation, and lighting. While classical computer vision techniques provided a foundation, advancements in deep learning (DL), especially CNNs, have significantly improved detection performance. Modern methods use hierarchical representations, enabling object detection in complex environments with occlusions and varying scales. Although many studies have reviewed specific deep learning models or object detection applications, few provide a comprehensive overview of both classical computer vision techniques and CNN-based approaches. This paper addresses this gap by offering an analysis of both. Key contributions include: 1. A review of classical computer vision techniques for object detection. 2. An analysis of general region proposal generation techniques. 3. A detailed review of convolution-based models for object detection, including two-stage and one-stage detectors. The paper is organized as follows: Section 2 covers classical computer vision techniques for object detection, Section 3 discusses region proposal generation, Section 4 explores CNN-based detection architectures, Section 5 reviews applications, Section 6 lists popular datasets, Section 7 covers evaluation metrics, and Section 8 concludes with future directions. II CLASSICAL COMPUTER VISION TECHNIQUES FOR OBJECT DETECTION Earlier computer vision techniques for image processing, particularly image similarity, relied on feature-based methods [ 3 , 4 , 5 , 6 , 7 , 8 ] . These methods focused on extracting distinctive image features to reduce computational costs while enabling robust image matching despite transformations like scaling or rotation [ 3 ] . The Scale-Invariant Feature Transform (SIFT) algorithm overcame the challenge of scaling by extracting features invariant to scale, rotation, brightness, and contrast [ 4 ] . Other feature extractors, like the Canny Edge Detector, contributed to tasks like image comparison and panoramic stitching by providing resilience to transformations and occlusions [ 5 ] . The Histogram of Oriented Gradients (HOG) technique enabled efficient image analysis by measuring gradient magnitudes and directions, creating descriptive feature vectors [ 6 ] . Traditional object detection involves three stages: 1. Proposal Generation: Scanning the image at various positions and scales to generate candidate bounding boxes, often using methods like sliding windows or selective search algorithms. 2. Feature Extraction: Extracting features from the identified regions to capture relevant visual patterns. 3. Classification: Classifying the extracted features using machine learning algorithms, such as support vector machine (SVM). In 2001, Viola et al. introduced a real-time (webcam based) facial detection classifier [ 7 ] . In 2005, Dalal et al. introduced an object detector using HOG features and an SVM classifier, effective across scales but limited by pose variations [ 6 ] . In 2009, Felzenszwalb et al. improved this with the Deformable Part Model (DPM), allowing flexible parts to handle poses, though it struggled with overlapping parts in multi-person images [ 8 ] . Studies from 2008 to 2012 on popular object detection datasets (see Section 5) showed key limitations in traditional methods. For instance, sliding windows require substantial computational resources and can generate redundant detections. Additionally, the performance of the classifier greatly impacts the results, necessitating more robust approaches. III GENERIC REGION PROPOSAL GENERATION TECHNIQUES Object detection models integrate a bounding box regressor within the classification network to accurately locate objects [ 9 ] . Traditionally, this involves feeding cropped images to the localization network, resulting in excessive inputs. An OverFeat model enhances efficiency by using a sliding window detector within convolution layers, scanning images with a large filter and stride [ 10 ] . However, indiscriminate scanning of background regions necessitates predicting potential object locations. Methods such as interest point detection, multiscale saliency, color contrast, edge detection, and super-pixel clustering are employed for this purpose [ 11 , 12 , 13 , 14 ] . For instance, multiscale saliency leverages the Fast Fourier Transform to analyze features at multiple scales [ 11 ] ; color contrast relies on color intensity differences [ 12 ] ; edge detection identifies edges, followed by density analysis [ 13 ] ; and super-pixel clustering groups similar pixels for detailed analysis [ 14 ] . Each method has specific limitations: multiscale saliency struggles with low-contrast objects, color contrast is ineffective with minimal contrast, edge detection may produce false positives or negatives, and super-pixel clustering requires refinement. Consequently, hybrid models are often developed to improve region proposal accuracy. IV CONVOLUTION BASED OBJECT DETECTION MODELS Object detection initially relied on manual feature design, focusing on patterns and edges. With CNN advancements, networks such as Visual Geometry Group Network (VGGNet) [ 15 ] and AlexNet [ 16 ] now autonomously extract features through convolution and pooling layers, with fully connected (FC) layers followed by a SoftMax layer for classification. For localization, the final FC layer outputs bounding box coordinates, unlike classical methods which use filters and machine learning-based models (e.g., SVMs). Training CNN-based models involve adjusting weights via backpropagation to align predictions with ground truth bounding boxes. Detection models fall into two categories: