David Moloney

3D Object Recognition Based on Volumetric Representation Using Convolutional Neural Networks

Following the success of Convolutional Neural Networks on object recognition and image classification using 2D images; in this work the framework has been extended to process 3D data. However, many current systems require huge amount of computation cost for dealing with large amount of data. In this work, we introduce an efficient 3D volumetric representation for training and testing CNNs and we also build several datasets based on the volumetric representation of 3D digits, different rotations along the x, y and z axis are also taken into account. Unlike the normal volumetric representation, our datasets are much less memory usage. Finally, we introduce a model based on the combination of CNN models, the structure of the model is based on the classical LeNet. The accuracy result achieved is beyond the state of art and it can classify a 3D digit in around 9 ms.

Object recognition speed improvement using BITMAP-HoG

Commonly, HoG/SVM classifier uses rectangular images for HoG feature descriptor extraction and training. This means significant additional work has to be done to process irrelevant pixels belonging to the background surrounding the object of interest. While some objects may indeed be square or rectangular, most of objects are not easily representable by simple geometric shapes. In Bitmap-HoG approach we propose in this paper, the irregular shape of object is represented by a bitmap to avoid processing of extra background pixels. Bitmap, derived from the training dataset, encodes those portions of an image to be used to train a classifier. Experimental results show that not only the proposed algorithm decreases the workload associated with HoG/SVM classifiers by 75% compared to the state-of-the-art, but also it shows an average increase about 5% in recall and a decrease about 2% in precision in comparison with standard HoG.

Speed improvement of object recognition using Boundary-Bitmap of histogram of oriented Gradients

Commonly, HoG/SVM classifier uses rectangular images for HoG feature descriptor extraction and training. This means that significant additional work has to be done to process irrelevant pixels belonging to the background surrounding the object of interest. Moreover, some areas of the foreground also can be eliminated from the processing to improve the algorithm speed and memory wise. In Boundary-Bitmap HoG approach proposed in this paper, the boundary of irregular shape of the object is represented by a bitmap to avoid processing of extra background and (partially) foreground pixels. Bitmap, derived from the training dataset, encodes those portions of an image to be used to train a classifier. Experimental results show that not only the proposed algorithm decreases the workload associated with HoG/SVM classifiers by 92.5% compared to the state-of-the-art, but also it shows an average increase about 6% in recall and a decrease about 3% in precision in comparison with standard HoG.

MvEcho - acoustic response modelling for auralisation

•Auralisation is the process of simulating the listening experience at a given position. •While many auralisation algorithms currently exist, the majority require a manual description of the environment. •In contrast, MvEcho removes the need for this manual description to provide an environment independent algorithm to approximate auralisation. •Initial implementation of MvEcho concerned the acoustic response of objects, namely the pyramid El Castillo, (shown below in Fig. 1) and has now been extended to model the response of rooms. •The algorithm is fully functional in C and is currently being ported to the Myriad 2.

Convolutional Neural Network for 3D object recognition using volumetric representation

Following the success of Convolutional Neural Networks (CNNs) on object recognition using 2D images, they are extended in this paper to process 3D data. Nearly most of current systems require huge amount of computation for dealing with large amount of data. In this paper, an efficient 3D volumetric object representation, Volumetric Accelerator (VOLA), is presented which requires much less memory than the normal volumetric representations. On this basis, a few 3D digit datasets using 2D MNIST and 2D digit fonts with different rotations along the x, y, and z axis are introduced. Finally, we introduce a combination of multiple CNN models based on the famous LeNet model. The trained CNN models based on the generated dataset have achieved the average accuracy of 90.30% and 81.85% for 3D-MNIST and 3D-Fonts datasets, respectively. Experimental results show that VOLA-based CNNs perform 1.5x faster than the original LeNet.

Efficient winograd-based convolution kernel implementation on edge devices

The implementation of Convolutional Neural Networks on edge Internet of Things (IoT) devices is a significant programming challenge, due to the limited computational resources and the real-time requirements of modern applications. This work focuses on the efficient implementation of the Winograd convolution, based on a set of application-independent and Winograd-specific software techniques for improving the utilization of the edge devices computational resources. The proposed techniques were evaluated in Intel/Movidius Myriad2 platform, using 4 CNNs of various computational requirements. The results show significant performance improvements, up to 54%, over other convolution algorithms.

Smart Doll: Emotion Recognition Using Embedded Deep Learning

Computer vision and deep learning are clearly demonstrating a capability to create engaging cognitive applications and services. However, these applications have been mostly confined to powerful Graphic Processing Units (GPUs) or the cloud due to their demanding computational requirements. Cloud processing has obvious bandwidth, energy consumption and privacy issues. The Eyes of Things (EoT) is a powerful and versatile embedded computer vision platform which allows the user to develop artificial vision and deep learning applications that analyse images locally. In this article, we use the deep learning capabilities of an EoT device for a real-life facial informatics application: a doll capable of recognizing emotions, using deep learning techniques, and acting accordingly. The main impact and significance of the presented application is in showing that a toy can now do advanced processing locally, without the need of further computation in the cloud, thus reducing latency and removing most of the ethical issues involved. Finally, the performance of the convolutional neural network developed for that purpose is studied and a pilot was conducted on a panel of 12 children aged between four and ten years old to test the doll.

Automatic Tree Annotation in LiDAR Data

LiDAR provides highly accurate 3D point cloud data for a number of tasks such as forest surveying and urban planning. Automatic classification of this data, however, is challenging since the dataset can be extremely large and manual annotation is labour intensive if not impossible. We provide a method of automatically annotating airborne LiDAR data for individual trees or tree regions by filtering out the ground measurements and then using the number of returns embedded in the dataset. The method is validated on a manually annotated dataset for Dublin city with promising results.

Face detection speed improvement using bitmap-based Histogram of Oriented gradien

Since the Viola-Jones seminal work, the boosted cascade with simple features has become the most popular and effective approach for practical face detection. More improved face detectors that can handle uncontrolled face detection scenarios have achieved by applying more advanced features such as Histogram of oriented Gradients (HoG). The great improvement in accuracy delivered by these methods has been accompanied by a large increase in the computational burden, which limited adoption in embedded solutions particularly. The improved bitmap-based HoG approaches resolved this problem by limitation of HoG window to non-rectangular irregular pattern of the object and its boundary avoid processing of extra background and (partially) foreground pixels respectively. In this paper, bHoG and bbHoG along with three different bitmap patterns are applied to the face detection problem to not only benefits from the robustness of HoG, but also to amend its high computational cost significantly. Experimental results show an decrease of 92.5% in the workload associated with HoG/SVM classifiers compared to the state-of-the-art, along with approximately the same performance as standard HoG and an average decrease about 5% in recall and precision in comparison for the smaller cell sizes.

Fully convolutional denoising autoencoder for 3D scene reconstruction from a single depth image

In this work, we propose a 3D scene reconstruction algorithm based on a fully convolutional 3D denoising autoencoder neural network. The network is capable of reconstructing a full scene from a single depth image by creating a 3D representation of it and automatically filling holes and inserting hidden elements. We exploit the fact that our neural network is capable of generalizing object shapes by inferring similarities in geometry. Our fully convolutional architecture enables the network to be unconstrained by a fixed 3D shape, and so it is capable of successfully reconstructing arbitrary scene sizes. Our algorithm was evaluated on a real word dataset of tabletop scenes acquired using a Kinect and processed using KinectFusion software in order to obtain ground truth for network training and evaluation. Extensive measurements show that our deep neural network architecture outperforms the previous state of the art both in terms of precision and recall for the scene reconstruction task. The network has been broadly profiled in terms of memory footprint, number of floating point operations, inference time and power consumption in CPU, GPU and embedded devices. Its small memory footprint and its low computation requirements enable low power, memory constrained, real time always-on embedded applications such as autonomous vehicles, warehouse robots, interactive gaming controllers and drones.