Object detection, recognition and classification using computer vision and artificial intelligence approaches
Year: 2018
Type: dissertation
Abstract: Object detection and recognition has been used extensively in recent years to solve numerus challenges in different fields.Due to the vital roles they play, object detection and recognition has enabled quantum leaps in many industry fields by helping to overcome some serious challenges and obstacles.For example, worldwide security concerns have drawn the attention and stimulated the use of highly intelligent computer vision technology to provide security in different environments and in diverse terrains.In addition, some wildlife is at present exposed to danger and extinction worldwide.Therefore, early detection and recognition of potential threats to wildlife have become essential and timely.The extent of using computer vision and artificial intelligence to convert the seemingly insecure world to a more secure one has been widely accepted.Such technologies are used in monitoring, tracking, organising, analysing objects in a scene and for a number of other countless purposes.With the goal of further expanding the application domains of computer vision and artificial intelligence, a novel algorithm to classify mixed industrial airborne waste particles is developed under the research context of this thesis.Initially, a dataset of industrial airborne particles was collected and imaged for this purpose.All images were captured using an advanced microscopic camera.The particles are segmented from the background image using K-means clustering and Hue, Saturation and Value (HSV) colour space features.Subsequently, the Hue, Saturation and Value (HSV) colour histogram data and Gray Level Co-occurrence Matrix (GLCM) texture features are combined into one feature vector of size 278 features, for use in dust sample classification.A feature selection algorithm is employed to minimise the number of features to ten, without compromising the accuracy that can be obtained when using all features.This significant minimisation of features used in classification would improve the processing time.A further computer vision system proposed in this thesis is for automated road surface type recognition aimed at use within autonomous vehicles.Traditional feature driven machine learning based algorithm and a deep learning based approach are proposed to recognise 5 different road surfaces namely: asphalt, concreate, rough, blocks and sand.Firstly, a novel algorithm is developed to exclude irrelevant objects and to create the region of interest within a snapshot image of the road surface.This segmentation algorithm eliminates unnecessary objects being considered for processing as the road in which the vehicle is being driven, by dividing the input image into top and bottom parts, to attempt a fast, region-dependent segmentation.Subsequently, the algorithm uses regiongrowing to create region of interests.For the recognition task, a feature based algorithm that uses a combination of colour histogram, Gray Level Co-occurrence Matrix (GLCM), Local Binary Pattern Feature Selection ...........
Author: Gharib Ismail Al Matroushi
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2892376930