1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Title:	Pembangunan kaedah pengecaman kejadian ragut menggunakan teknik aliran optik
Authors:	Norazlin Ibrahim (P40846)
Supervisor:	Mohd Marzuki Mustafa, Prof. Ir. Dr.
Keywords:	Pengecaman kejadian ragut Aliran optik Video surveillance
Issue Date:	7-Oct-2014
Description:	Jenayah ragut yang semakin berleluasa telah mencetuskan kebimbangan umum. Sebagai langkah membenteras jenayah ini, pihak berkuasa tempatan telah memasang sistem pengawasan video (SPV) di beberapa lokasi tertentu. Namun begitu, SPV sedia ada kurang berkesan dalam membenteras jenayah khususnya ragut kerana SPV sedia ada lebih cenderung digunakan untuk merakam aktiviti di lokasi tertentu dan memantau secara insani. Dalam kebanyakan kes, hasil rakaman video hanya digunakan sebagai bahan siasatan pasca kejadian. Justeru, suatu SPV yang lebih cekap adalah diperlukan supaya kejadian jenayah ragut khususnya dapat dikenal pasti secara automatik, pantas dan cekap. Salah satu teknik yang boleh digunakan adalah teknik aliran optik (TAO) yang biasa digunakan dalam kajian pengesanan pergerakan objek. TAO sesuai kerana keupayaannya mempamer corak pergerakan objek yang dapat membantu mengesan perubahan pergerakan mendadak yang boleh dikaitkan dengan aktiviti jenayah ragut. Oleh itu, objektif kajian adalah untuk membangunkan suatu kaedah pengecaman perubahan pergerakan mendadak objek yang boleh dikaitkan dengan jenayah ragut. Kajian ini memfokus kepada TAO yang dijadikan asas untuk proses sarian vektor sifat untuk tujuan pengecaman. TAO yang melibatkan algoritma Horn-Schunk, Lucas-Kanade dan Brox telah dipertimbangkan dalam ujikaji awal untuk menentukan TAO yang terbaik. Selepas itu, vektor sifat disari terus dari hasil TAO untuk mendapat ciri jumlah magnitud aliran optik (JMAO) dan ciri taburan sudut (TS). Fasa pengesanan pula melibatkan tiga kaedah pengesan kerangka imej yang mengandungi perubahan pergerakan objek secara mendadak. Kaedah titik ambang, penuras Kalman dan ukuran kemiripan (UK) yang melibatkan ciri JMAO dan ciri TS digunakan untuk pengesanan kerangka imej mula dan akhir. Setelah kerangka imej yang mengandungi perubahan objek secara mendadak ditentukan, kerangka tersebut dikelaskan untuk menentukan samada pergerakan objek adalah jenayah ragut atau tidak. Tiga jenis pengelas telah dipertimbangkan iaitu Pengelas Naive Bayes, Tree Bagger dan Mesin Penyokong Vektor (SVM) untuk tujuan pengecaman kejadian jenayah ragut. Keputusan awal kajian menunjukkan algoritma TAO yang terbaik adalah algoritman Horn-Schunk. Justeru, algoritma Horn-Schunk telah digunakan secara menyeluruh dalam membangunkan kaedah pengecaman jenayah ragut melalui pengesanan perubahan pergerakan secara mendadak pada kerangka. Empat skema gabungan vektor sifat ciri JMAO, ciri TS dan teknik pengesanan kerangka berbeza telah dibangunkan. Hasil pengecaman terbaik diperoleh melalui teknik pengesanan kerangka imej yang melibatkan gabungan pengesanan kerangka menggunakan penuras Kalman dan vektor sifat ciri TS. Peratus kejituan apabila diuji dengan pengelas Tree Bagger, dan pengelas SVM adalah masing-masing 90% dan 86%. Secara ringkas, gabungan kaedah pengesanan kerangka menggunakan penuras Kalman dan vektor sifat ciri TS dan juga pengelas Tree Bagger adalah kaedah terbaik.,The increasingly rampant snatch theft activities have triggered public concern. As a preventive action, several local authorities had installed a video surveillance system (VSS) in selected locations. However, the current VSS is ineffective in reducing snatch theft activities since it is more likely to be used for manual monitoring and recording of events. In most cases, the recorded videos are only used as proof for post incident. Hence, a VSS that is more efficient is needed so that snatch theft activities can be identifiable automatically, fast and efficiently. One of the techniques that can be used is the optical flow technique (OFT) that is commonly used in object movement detection study. OFT is suitable due to its ability to display the object movement pattern that can help detect sudden change of movement that can be associated with snatch theft crime activity. As such, the objective of the study is to develop a method to recognize sudden change of object movement that can be associated with the crime. This study mainly focuses on the use of OFT to extract feature vectors for the snatch theft event recognition purpose. Earlier in this research, Horn-Schunk, Lucas- Kanade and Brox algorithms of the OFT were considered to determine the best OFT algorithm. Next, the feature vectors are directly extracted from the OFT output. The feature vectors are the sum of Optical Flow Magnitud (JMAO) and the angular distribution (TS). The detection phase involves method to determine the image frame that contains sudden movement of an object. For determining the starting and ending frame consisting of the sudden change in object movement, the threshold method, Kalman filtering and similarity measure along with the JMAO and TS feature vectors are used. Upon detecting the frame that contain sudden movement, the frame is then classified as either a frame with snatch theft event or otherwise. Three different classifiers were considered for the recognition task of identifying the frame that consists of a snatch theft event. The classifiers are the Naive Bayes classifier, the Tree Bagger classifier and the Support Vector Machine. In the preliminary results, the best OFT algorithm was Horn-Schunk and hence, it was used entirely for the development of the snatch theft recognition method via sudden event recognition. Four different schemes comprising combinations of the JMAO and TS features and four different frame detection techniques were developed and tested. The best recognition was obtained using the frame detection technique which involves the use of Kalman filter and the TS feature vector. This combination when tested with the Tree Bagger and SVM classifier produce the best results with accuracies of 90% and 86%, respectively. In short, the best method for the snatch theft recognition is using the combination of Kalman filter for frame detection, the TS feature vectors and Tree Bagger classifier.,PhD
Pages:	183
Call Number:	TK6680.3.N646 2014 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Engineering and Built Environment / Fakulti Kejuruteraan dan Alam Bina

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.