2020 Seminars – Centre for Robotics & Vision @ AUT

For attending CeRV research seminars 2020, just come and sit in; those are public. For date, time, and venue, please see at each individual seminar.

10 September 2020 -03 December 2020 (Thursday, Online) Topics on “Various Methods for Deep Learning and Intelligent Surveillance”
(1) Human behaviour recognition using deep learning (Jia Lu, 12.00pm, 10 September 2020)
(2) Vehicle-related scene understanding using CapsNets (Xiaoxu Liu, 12.00pm, 17 September 2020)
(3) Assuring privacy-preservation in mining medical text materials for COVID-19 cases – A natural language processing perspective (Bo Ma, 12.00pm, 24 September October 2020)
(4) Apple ripeness identification (Bingjie Xiao, 12.00pm, 1 Oct 2020)
(5) Computational methods of visual blockchain for smart city (Kasun Tharaka, 12.00pm, 8 oct 2020)
(6) Deep learning methods for human action recognition (Zeqi Yu, 12.00pm, 15 OCT 2020)
(7) Fruit detection using CenterNet (Kun Zhao, 11.30am, 22 Oct 2020)
(8) Improving security for video watermarking (Milan Gupta, 12.00pm, 29 Oct 2020 )
(9) Colorising grayscale CT images of human lungs using deep learning method (Yuewei Wang, 12.00pm, 5 Nov 2020)
(10) Traffic sign recognition using deep learning (Zhongbing Qin, 12.00pm, 12 Nov 2020)
(11) Visual object detection for tree leaves based on deep learning (Lei Wang, 12.00pm, 19 Nov 2020)
(12) Auto speech recognition using deep learning (Sendong Liang, 12.00pm, 26 Nov 2020)
(13) Sheep counting (Wenjun Liu, 12.00pm, 03 Dec 2020)
(14) LiDAR and vision sensor fusion using deep neural networks for road scene perception in autonomous vehicles (Sabeeha Mehtab, 12.00pm, 10 Dec 2020)
02 July 2020 -03 September 2020 (11.00am, Thursday, Online) Topics on “CCV: Concise Computer Vision.”
(1) Image Data (Wei Qi Yan, 02 July 2020)
(2) Image Processing (Wei Qi Yan, 09 July 2020)
(3) Image Analysis (Wei Qi Yan, 16 July 2020)
(4) Dense Motion Analysis (Wei Qi Yan, 23 July 2020)
(5) Image Segmentation (Wei Qi Yan, 30 July 2020)
(6) Cameras, Coordinates, and Calibration (Wei Qi Yan, 06 August 2020)
(7) 3D Shape Reconstruction (Wei Qi Yan, 13 August 2020)
(8) Stereo Matching (Wei Qi Yan, 20 August 2020)
(9) Feature Detection and Tracking (Wei Qi Yan, 27 August 2020)
(10) Object Detection (Wei Qi Yan, 03 September 2020)
28 May 2020 -16 July 2020 (12.00pm, Thursday, Online) Topics on “Postgraduate Mathematics and Deep Learning.”
(1) Functional Analysis (Wei Qi Yan, 28 May 2020)
(2) Finite Fields (Wei Qi Yan, 04 Jun 2020)
(3) Probabilistic Graphical Models (Wei Qi Yan, 11 Jun 2020)
(4) Reinforcement Learning (Wei Qi Yan, 18 Jun 2020)
(5) Generative Adversarial Networks (Wei Qi Yan, 25 Jun 2020)
(6) Manifold Learning (Wei Qi Yan, 02 July 2020)
(7) Autoencoder (Wei Qi Yan, 09 July 2020)
(8) CapsNets (Wei Qi Yan, 16 July 2020)
30 January — 7 May 2020 (12.00pm, Thursday) Topics on “Applications of Deep Learning.”
(1) Currency Series Number Recognition Using Deep Learning (Xin Ma, 30 January 2020)
(2) Human Behaviour Recognition Using Deep Learning (Jia Lu, 06 February 2020)
(3) Diagnosis of Alzheimer’s Disease Using Deep Learning (Shouming Sun, 13 February 2020)
(4) Vehicle-Related Scene Understanding Using Deep Learning (Xiaoxu Liu, 20 February 2020)
(5) Fruits Freshness Grading Using Deep Learning (Yuhang Fu, 27 February 2020, Online)
(6) Anomaly Recognition Using Deep Learning (Na An, 05 March 2020, Online)
(7) Fruits Ripeness Recognition Using Deep Learning (Bingjie Xiao, 12 March 2020, Online)
(8) Image Salience Detection Using Deep Learning (Jianfeng Liu, 19 March 2020, Online)
(9) Multiple Virus Recognition Using Deep Learning (Luxin Zhang, 26 March 2020, Online)
(10) Coins Recognition Using Deep Learning (Yufeng Xiang, 02 April 2020, Online)
(11) Fruits Classification Using Deep Learning (Kun Zhao, 09 April 2020, Online)
(12) Symbolic Computations Using Reinforcement Learning (Nikhil Kumar Yadav, 16 April 2020, Online)
(13) Object Detection Using Manifold Learning (Saurabh Inge, 23 April 2020, Online)
(14) Human Action Analysis Based on Videos (Zeqi Yu, 30 April 2020, Online)
(15) Evaluations of Visual Watermarking (Milan Gupta, 7 May 2020, Online)

10 September 2020 (12.00pm, Thursday, Online) Jia Lu (Computer AUT) Topic on “Human Behaviour Analysis Using Deep Learning”
In order to achieve sufficient recognition accuracy, both spatial and temporal information was acquired and used to implement human behaviour analysis. We propose a novel Selective Kernel Network (SKNet) with attention mechanism, which yields encouraging results for real-time human behaviour recognition. The key contributions of this paper are: (1) The SKNet with attention mechanism that achieves the best accuracy of human behaviour recognition at the rate up to 98.7% overall, based on various public datasets. (2) YOLOv4 + LSTM network which is able to achieve 97.87% of total accuracy based on our own dataset, using spatial-temporal information extracted from our recorded video footages.
09 July 2020 (1.00pm, Thursday, Online) Xiaoxu Liu (Computer Science, AUT) Topic on “Vehicle-Related Scene Understanding Using Deep Learning”
Automated driving technology is an inevitable trend in the future development of transportation, it is also one of the eminent achievements in the matter of artificial intelligence. In this talk, we implement an efficient and robust scene segmentation model using capsule network as basic framework for detailed preservation. We collected a large amount of image data of Auckland traffic scenes on the highway and labelled these data for classification. There are six classes of labels in the proposed model and 1,200 image data are employed for model training. The proposed model has compared with other advanced models to evaluate its effectiveness and robustness.
14 May 2020 (12.00pm, Thursday, Online) Alex Zhu (Computer Science, AUT) Topic on “Return-Oriented Programming: Attack, Evasion, and Defence”
Most of existing Intrusion Detection Systems (IDS) / Intrusion Prevention Systems (IPS) for cloud computing cannot effectively defend Return-Oriented Programming (ROP) attacks which apply code reusing and exploiting technique without the need for code injection. We analyse the collected data to find a research problem and propose a solution for resolving the problem. We improve cloud computing security and remove the obstacles. The work in this talk explores our defence against security attacks in cloud computing, which could benefit all cloud computing shareholders and promote cloud development.
30 April 2020 (12.00pm, Thursday, Online) Bob Ma (Computer Science, AUT) Topic on “Privacy-Preserving Federated Machine Learning Methods”
The recent developed generative adversarial network (GAN) based attacks demonstrate that simply maintaining data locality during training processes does not provide sufficient privacy guarantees. To tackle above problems, this research project focuses on: (1) To develop a new dynamic entropy-based noise generation method with the differential privacy for betterment the privacy protection of the federated machine learning architecture. (2) To develop a new distributed stochastic gradient descent (SGD) algorithm for betterment the learning performance of the federated a machine learning architecture. (3) To develop a new generative adversarial network (GAN) based approach to measure the privacy protection level of output data from the federated machine learning architecture.
27 February 2020 (2.00pm, Thursday, WZ630) Zahra Moayed (EEE, AUT) Topic on “Automated Multi-view Safety Analysis at Complex Road Intersections”
The safety of pedestrians and vehicles at traffic intersections is a major concern for transport practitioners these days due to the high number of reported accidents and fatalities. In this talk, an automatic vision-based system is used to understand traffic patterns and to analyse participants’ safety at intersections. The major novelty of the algorithms is to present a robust safety analyser using four calibrated cameras at a real intersection. To inspect safety, the characteristics of the participants are extracted; detection, classification, and tracking use a fusion of appearance-based and motion-based methods. Deep learning proves its ability to take part at this stage, handling tradeoffs among accuracy, time sufficiency, and robustness, while being associated with motion parameters. The talk is further extended to consider the safety measurements and interaction risk factors so as to analyse the potential risks for each participant in the form of a single value.
16 January 2020 (10.00am, Thursday, WZ630) Yue Ding (CS, AUT) Topic on “Enhanced Reasoning and Learning Algorithms for General Game Playing”
In terms of reasoning techniques, Prolog and Proposition Network (PropNet) are two effective methods integrated in many GDL reasoners. As playing strategies, reinforcement learning (RL) (e.g., Q-Learning) is applied widely to predict long-term rewards and make selections. Additionally, deep neural networks trained by RL (DRL) is verified to be advisable in more general environments with less domain knowledge.

16 January 2020 (1.00pm, Thursday, WZ630) Sagar Dake (CS, AUT) Topic on “Object Detection and Classification on 360-Degree Cameras Feed using Deep Learning”
The information about 2/3D data is necessary for navigation and control applications. In this research work, we have detected moving objects with the integration of additional prediction layers into the conventional YOLO v3 network. We increased the accuracy of blurry image or poor video quality from 360-degree cameras that detects the overlapped objects with insufficient quality input. Our initially proposed detector achieves 85.29% Mean Average Precision (MAP). We achieved this precision and accuracy based on the UA-DETRAC dataset of 360-degree video cameras by using various models.