Feature selection using Ant Colony Optimization (ACO) and Road Sign Detection and Recognition (RSDR) system
Introduction
Recently there is an evident surge in the use of unmanned automatic driving technology (Sezer et al., 2011). This surge has gained quite a bit of attention from all over the world right from researches to industry communities. In this particular traffic signs recognition and classification play have an integral part in determining related aspects. Significant research and learning have been directed towards resolving inherent issues. Issues related to insufficient illumination, partial occlusion and serious deformation have raised many impediments during traffic detection (de la Escalera et al., 2004, de la Escalera et al., 1997).
Most of the researched and published work depicts results from a two-stage sequential approach taken that involves two stages, detection and recognition (or classification). These are used to identify traffic signs drawn from a series of captured roadside images. Main focus here during detection is identifying an interest region within the image and simultaneously ensuring region of interest has a traffic sign. Recognition is carried out to identify traffic sign that characteristically unique during process of detection.
Detection of sign candidates is possible here when information pertaining to the color has been deployed (Piccioli, De Micheli, Parodi, & Campani, 1996), (Gao, Podladchikova, Shaposhnikov, Hong, & Shevtsova, 2006), followed by geometrical edge (Garcia-Garrido, Sotelo, & Martin-Gorostiza, 2006) or corner analysis. Though, a few other authors choose a strictly colourless approach. This is also because they do not take in to consideration colour segmentation as being completely reliable on account of its sensitivity to several other related factors which include several factors like weather conditions, reflectance of the sign’s surface, distance calculated from target sign, or time of day.
Colour information in the studies carried out during detection, majority are non-Red Green Blue (RGB) color spaces based. The Hue-Saturation-Value (HSV) color model has been one such prominent model deployed as using human color perception is the primary determining factor. Additionally, this model is relatively invariant with respect to illumination changes. HSV model helps to classify test sign images as distinctly different categories. This approach is extremely well suited as it uses colour-based detection and segmentation of road signs using IHLS colour space which has already been suggested (Fleyeh, 2004). Based on the studies presented, colour appearance models selected were found to be the obvious reasonable choices in comparison with the standard Red Green Blue (RGB) model.
A pixel-based approach deployed during recognition stage as well as detected sign of the class comprehended using a cross-correlation template with a corresponding and apt technique (Fleyeh, 2004). The other popular approach that has been used is the feature-based earmarked with a 95% success recognition rate as deduced from experiments that use still camera images. Though, previous studies aimed at narrowing down traffic signs or situations subsets. However, several authors have deployed a single semantic category for testing that includes speed limit signs, or relatively dissimilar signs from many categories. The selection of features from the road sign images becomes very difficult task.
Feature selection is an integral function in many pattern recognition related issues (Guyon & Elisseeff, 2003); though many aspects maybe considered while describing an image, only a few are prolific and persuasive during the process of classification. The more the number of features in no way indicates a superior grouping execution. However in lieu of this it includes determination that is carried out to select a smaller and applicable component subset while focussing on diminishing the dimensionality of feature space, which subsequently will facilitate arrangement precision and reduce time utilization. Using feature selection algorithms are used primarily while resolving machine learning problems that involve multiple digital images. Digital images here are selected to facilitate image related aspects which include classification, enhancement, compression and segmentation. Every advanced image has varying power esteems pixels that have been set in an example.
Identification of objects that are distinctive can be easily done by anyone’s naked eyes without missing any trademark features which include colour, shape, geometry or texture features. Where object’s new subset is apparent to any individual, the related protest winds up inevitably detect the same as the brain triggers information and begins to investigate new subset bearing in mind current features as a guideline. Machine learning necessitates a list of capabilities that bear clearly distinctive trademark features and these maybe premium, advanced images or can be compared to any other measurable object. Trademark feature set of objects like these is evaluated and focuses on determination and classification. The generic procedure deployed for feature selection includes 4 important steps as illustrated in Fig. 1. Step 1 - Feature Subset Selection, Step 2 - Feature Subset Evaluation, Step 3 - Stopping Criteria and Step 4 - Result Validation.
Road sign detection and recognition system are laborious vision tasks. Various important issues need to be resolved. Introduction of an efficient RSDR system which easily includes pre-processing, edge detection, feature extraction, features selection and ensemble Fuzzy SVM (EFSVM) classifier has been the main objective in this study. This chapter mainly covers the RSDR system. The datasets such as German Traffic Signs Recognition Benchmark (GTSRB) and German Traffic Signs Detection Benchmark (GTSDB) are used for RSDR analysis.
Section snippets
Literature review
A lot of research has been focussed towards this issue of finding the color invariance. Gomez-Moreno et al., 2010, Bui-Minh et al., 2012, Maldonado-Bascon et al., 2007, and Fleyeh (2006) explained that the choice of a suitable colour space is very important and that it plays a very important role in providing an accurate detection result.
Ruta, Li, and Liu (2010) demonstrated that a series of thresholds becomes difficult to control as when light and other factors have a major impact that finally
Road sign detection and recognition system
A unique but practically renowned RSDR system has been suggested and this includes pre-processing, feature extraction, features selection and ensemble FSVM (EFSVM) classifier. The main aim of this work will be primarily focus on the RSDR system shown in Fig. 2.
Experimentation results
Experiments that have been conducted here are solely for the purpose of identifying road sign contents on the basis of segmented regions of the road signs. Basic shape and colour as mentioned earlier are taken into consideration. The experimental work for the proposed scheme RSDR is done by utilizing the datasets such as German Traffic Signs Recognition Benchmark (GTSRB) and German Traffic Signs Detection Benchmark (GTSDB). These Datasets are freely available in the Institute for
Conclusion and future work
It has been our earnest endeavour to present a paper that is innovative and offers a unique perspective to novel traffic RSDR system. The system includes the following: (i) pre-processing, edge detection, feature extraction, feature selection and, (iv) ESVM classifiers. For the system that has been proposed here, primarily traffic signs have undergone the process of color conversion and noise removal executed by deploying Gaussian filtering. Secondly edge detection is performed using canny edge
Conflict of interest
The authors declared that there is no conflict of interest.
References (22)
Ant colony optimization: Introduction and recent trends
Physics of Life Reviews
(2005)- et al.
Recognition of traffic signs based on their colour andshape features extracted using human vision models
Journal of Visual Communication and Image Representation
(2006) - et al.
An improved ant colony algorithm for fuzzy clustering in image segmentation
Neurocomputing
(2007) - et al.
A review of ant algorithms
Expert Systems with Applications
(2009) - et al.
A robust method for road sign detection and recognition
Image and Vision Computing
(1996) - et al.
Real-time traffic sign recognition from video by class-specific discriminative features
Pattern Recognition
(2010) - et al.
Automatic road-sign detection and classification based on support vector machines and HOG descriptors
ISPRS Annals of Photogrammetry, Remote Sensing & Spatial Information Sciences
(2014) - Bui-Minh, T., Ghita, O., Whelan, P. F., & Hoang, T. (2012). A robust algorithm for detection and classification of...
- et al.
Visual sign information extraction and identification by deformable models for intelligent vehicles
IEEE Transactions on Intelligent Transportation Systems
(2004) - et al.
Road traffic sign detection and classification
IEEE Transactions on Industrial Electronics
(1997)
Cited by (43)
Feature selection using a sinusoidal sequence combined with mutual information
2023, Engineering Applications of Artificial IntelligenceTraffic sign recognition using proposed lightweight twig-net with linear discriminant classifier for biometric application
2023, Image and Vision ComputingDispersed foraging slime mould algorithm: Continuous and binary variants for global optimization and wrapper-based feature selection
2022, Knowledge-Based SystemsCitation Excerpt :In order to improve the performance of the credit scoring model, Jadhav et al. [50] investigated the genetic algorithm based feature selection (FS). Jayaprakash and KeziSelvaVijila [51] successfully performed feature selection by deploying the ant colony optimization algorithm (ACO) to identify the most prominent features. Jia et al. [52] showed three hybrid algorithms based on the seagull optimization algorithm (SOA) and thermal exchange optimization (TEO) to solve the feature selection problem.
Feature selection for sky image classification based on self adaptive ant colony system algorithm
2023, International Journal of Electrical and Computer Engineering