Monika Sharma

Road condition monitoring and alert application: Using in-vehicle Smartphone as Internet-connected sensor

The proposal describes a road condition monitoring and alert application using the in-vehicle Smartphone as connected sensors, which are connected to an Internet-of-Things platform over the Internet. In addition to providing a generic Internet-of-Things based platform, the proposed solution brings in novel energy-efficient phone-orientation-agnostic accelerometer analytics in phone, reduces the data volume that needs be communicated between phone and the back-end over Internet, brings in multi-user fusion concepts to create authentic road condition maps and addresses privacy concerns for the phone user for sharing the required data.

An AR Inspection Framework: Feasibility Study with Multiple AR Devices

We present an Augmented Reality (AR) based re-configurable framework for inspection that can be utilized in cross-domain applications such as maintenance and repair assistance in industrial inspection, health sector to record vitals, and automotive/avionics domain inspection, amongst others. The novelty of the inspection framework as compared to the existing counterparts are three fold. Firstly, the inspection check-list can be prioritized by detecting the parts viewed in inspectors field using deep learning principles. Second, the backend of the framework is easily configurable for different applications where instructions and assistance manuals can be directly imported and visually integrated with inspection type. Third, we conduct a feasibility study on inspection modes such as Google Glass, Google Cardboard, Paper based and Tablet for inspection turnaround time, ease, and usefulness by taking a 3D printer inspection use-case.

A density based method for automatic hairstyle discovery and recognition

This paper presents a novel method for discovery and recognition of hairstyles in a collection of colored face images. We propose the use of Agglomerative clustering for automatic discovery of distinct hairstyles. Our method proposes automated approach for generation of hair, background and face-skin probability-masks for different hairstyle category without requiring manual annotation. The probability-masks based density estimates are subsequently applied for recognizing the hairstyle in a new face image. The proposed methodology has been verified with a synthetic dataset of approximately thousand images, randomly collected from the Internet.

Histogram of gradient magnitudes: A rotation invariant texture-descriptor

We propose a new low-dimensional rotation invariant local texture-descriptor with a low computational complexity in this paper. Commonly used local texture features are sensitive to rotation of texture patterns in an image. On the other hand, rotation-invariant features are generally computationally expensive. The proposed feature descriptor is based on gradient of magnitude of pixel-intensities that is naturally rotation-neutral. Experiments with several datasets show better performance for the proposed texture-descriptor as compared to the existing texture-descriptors in image classification and segmentation tasks.

Crowdsourcing for Chromosome Segmentation and Deep Classification

Metaphase chromosome analysis is one of the primary techniques utilized in cytogenetics. Observations of chromosomal segments or translocations during metaphase can indicate structural changes in the cell genome, and is often used for diagnostic purposes. Karyotyping of the chromosomes micro-photographed under metaphase is done by characterizing the individual chromosomes in cell spread images. Currently, considerable effort and time is spent to manually segment out chromosomes from cell images, and classifying the segmented chromosomes into one of the 24 types, or for diseased cells to one of the known translocated types. Segmenting out the chromosomes in such images can be especially laborious and is often done manually, if there are overlapping chromosomes in the image which are not easily separable by image processing techniques. Many techniques have been proposed to automate the segmentation and classification of chromosomes from spread images with reasonable accuracy, but given the criticality of the domain, a human in the loop is often still required. In this paper, we present a method to segment out and classify chromosomes for healthy patients using a combination of crowdsourcing, preprocessing and deep learning, wherein the non-expert crowd from CrowdFlower is utilized to segment out the chromosomes from the cell image, which are then straightened and fed into a (hierarchical) deep neural network for classification. Experiments are performed on 400 real healthy patient images obtained from a hospital. Results are encouraging and promise to significantly reduce the cognitive burden of segmenting and karyotyping chromosomes.

Pre-trained classifiers with One Shot Similarity for context aware face verification and identification

Most affect based systems analyse facial expressions for emotion detection, and utilize face detection and recognition methods in order to do effective affect analysis. Recent work has demonstrated the efficacy of deep architectures for face recognition by training as classifiers on voluminous datasets. Some architectures are trained as classifiers, and some directly learn an embedding via a triplet loss function. In this paper, we consider the case of one shot prediction from the feature space learnt initially via classification, i.e. we consider the situation where we have a pre-trained model, but do not have access to the training data and are required to make predictions on novel faces with just one training image per identity. We utilize the one shot similarity metric in order to compute similarity scores and compare it with the state-of-the-art results on the Youtube videos face dataset (YTF). We demonstrate the effect of temporal context on frame wise face recognition, and use a probabilistic majority voting scheme over past frames to determine current frame identity. Additionally, we found a number of labelling errors in the Youtube face dataset that were not published in the errata, and have published the same online for the benefit of the community.

Automatic Container Code Recognition via Spatial Transformer Networks and Connected Component Region Proposals

Container identification and recognition is still performed manually or in a semi-automatic fashion in multiple ports globally. This results in errors and inefficiencies in port operations. The problem of automatic container identification and recognition is challenging as the ISO standard only prescribes the pattern of the code and does not specify other parameters such as the foreground and background colors, font type and size, orientation of characters (horizontal or vertical) so on. Additionally, the corrugated surface of container body makes the two dimensional projection of the text on three dimensional containers slanted and jagged. We propose a solution in the form of an end-to-end pipeline that uses Region Proposals generated based on Connected Components for text detection in conjunction with Spatial Transformer Networks for text recognition. We demonstrate via our experimental results that the pipeline is reliable and robust even in situations when the code characters are highly distorted and outperforms the state-of-the-art results for text detection and recognition over the containers. We achieve text coverage rate of 100% and text recognition rate of 99.64%.

Unsupervised Geo-Demographic Classification of City-Area Using Multimodal Multimedia Data

Geo-demographical segmentation of a city area is of interest in many social as well business contexts. In this paper, we present a new method for demographic segmentation of a city by analyzing satellite imagery and other forms of neighborhood data. We explore the feature-set that can be used for effective geo-demographic segmentation of satellite imagery. In particular, we study the effectiveness of a few color and texture features for the purpose. Our method essentially involves analysis of large volume of image data, for which we have devised a two-step process and approximate clustering methods. We have experimented with satellite images of two major cities to illustrate the results. We have integrated telecom data that signify human activity with satellite imagery to achieve enhanced segmentation of the demographic regions.