Search

In autonomous vehicles, depth information for the environment surrounding the vehicle is commonly extracted using time-of-flight (ToF) sensors such as LiDARs and RADARs. Those sensors have some limitations that may potentially degrade the quality and utility of the depth information to a substantial extent. An alternative solution is depth estimation from stereo pairs. However, stereo matching and depth estimation often fails at ill-posed regions including areas with repetitive patterns or textureless surfaces which are commonly found on planar surfaces. This paper focuses on designing an efficient framework for stereo depth estimation, using deep learning technique, that is robust against the mentioned ill-posed regions. With the observation that disparities of all pixels belonging to planar areas (scene plane) viewed by two rectified stereo images can be described using affine transformations, our proposed method predicts pixel-wise affine transformation parameters based on the depth information encoded in the aggregated cost volume. We also introduce a propagation term which enforces all pixels belonging to the same scene plane to be transformed using the same parameters. Disparity can then be computed by multiplying the predicted affine parameters with the corresponding pixel locations. The proposed method was evaluated on several benchmark datasets. We are able to obtain competitive results and at the same time reducing the processing time of common convolution neural network (CNN) in stereo matching by 50%. Analysis of the findings shows that our method can produce reliable results at the ill-posed regions which are challenging to the current state-of-the-arts methods.

This paper presents a new solution for multi-target tracking over a network of sensors with limited spatial coverage. The proposed solution is based on the centralized data fusion architecture. The main contribution of the paper is the introduction of a new track-to-track fusion approach in which the posterior distributions of multi-target states, reported by various sensor nodes, are fused in a way that the redundant information are combined and the rest complement each other. The proposed solution is formulated within the labeled random finite set framework in which the fused posterior incorporates all the state and label information provided by multiple sensor nodes. The performance of the proposed method is evaluated via simulation experiments that involve challenging tracking scenarios. The proposed method is implemented using sequential Monte Carlo method and the results confirm its effectiveness.

State-of-the-art stereo matching networks trained only on synthetic data often fail to generalize to more challenging real data domains. In this paper, we attempt to unfold an important factor that hinders the networks from generalizing across domains: through the lens of shortcut learning. We demonstrate that the learning of feature representations in stereo matching networks is heavily influenced by synthetic data artefacts (shortcut attributes). To mitigate this issue, we propose an Information-Theoretic Shortcut Avoidance~(ITSA) approach to automatically restrict shortcut-related information from being encoded into the feature representations. As a result, our proposed method learns robust and shortcut-invariant features by minimizing the sensitivity of latent features to input variations. To avoid the prohibitive computational cost of direct input sensitivity optimization, we propose an effective yet feasible algorithm to achieve robustness. We show that using this method, state-of-the-art stereo matching networks that are trained purely on synthetic data can effectively generalize to challenging and previously unseen real data scenarios. Importantly, the proposed method enhances the robustness of the synthetic trained networks to the point that they outperform their fine-tuned counterparts (on real data) for challenging out-of-domain stereo datasets., Comment: 11 pages, 4 figures. Accepted by CVPR2022

Retinal swelling due to the accumulation of fluid is associated with the most vision-threatening retinal diseases. Optical coherence tomography (OCT) is the current standard of care in assessing the presence and quantity of retinal fluid and image-guided treatment management. Deep learning methods have made their impact across medical imaging, and many retinal OCT analysis methods have been proposed. However, it is currently not clear how successful they are in interpreting the retinal fluid on OCT, which is due to the lack of standardized benchmarks. To address this, we organized a challenge RETOUCH in conjunction with MICCAI 2017, with eight teams participating. The challenge consisted of two tasks: fluid detection and fluid segmentation. It featured for the first time: all three retinal fluid types, with annotated images provided by two clinical centers, which were acquired with the three most common OCT device vendors from patients with two different retinal diseases. The analysis revealed that in the detection task, the performance on the automated fluid detection was within the inter-grader variability. However, in the segmentation task, fusing the automated methods produced segmentations that were superior to all individual methods, indicating the need for further improvements in the segmentation performance.

This paper presents a new approach to solve unsupervised video object segmentation~(UVOS) problem (called TMNet). The UVOS is still a challenging problem as prior methods suffer from issues like generalization errors to segment multiple objects in unseen test videos (category agnostic), over reliance on inaccurate optic flow, and problem towards capturing fine details at object boundaries. These issues make the UVOS, particularly in presence of multiple objects, an ill-defined problem. Our focus is to constrain the problem and improve the segmentation results by inclusion of multiple available cues such as appearance, motion, image edge, flow edge and tracking information through neural attention. To solve the challenging category agnostic multiple object UVOS, our model is designed to predict neighbourhood affinities for being part of the same object and cluster those to obtain accurate segmentation. To achieve multi cue based neural attention, we designed a Temporal Motion Attention module, as part of our segmentation framework, to learn the spatio-temporal features. To refine and improve the accuracy of object segmentation boundaries, an edge refinement module (using image and optic flow edges) and a geometry based loss function are incorporated. The overall framework is capable of segmenting and finding accurate objects' boundaries without any heuristic post processing. This enables the method to be used for unseen videos. Experimental results on challenging DAVIS16 and multi object DAVIS17 datasets shows that our proposed TMNet performs favourably compared to the state-of-the-art methods without post processing.

Consensus maximisation (MaxCon), which is widely used for robust fitting in computer vision, aims to find the largest subset of data that fits the model within some tolerance level. In this paper, we outline the connection between MaxCon problem and the abstract problem of finding the maximum upper zero of a Monotone Boolean Function (MBF) defined over the Boolean Cube. Then, we link the concept of influences (in a MBF) to the concept of outlier (in MaxCon) and show that influences of points belonging to the largest structure in data would generally be smaller under certain conditions. Based on this observation, we present an iterative algorithm to perform consensus maximisation. Results for both synthetic and real visual data experiments show that the MBF based algorithm is capable of generating a near optimal solution relatively quickly. This is particularly important where there are large number of outliers (gross or pseudo) in the observed data., To appear in CVPR 2021 as an ORAL paper. arXiv admin note: text overlap with arXiv:2005.05490

Prostate cancer (PCa) is the second most frequent type of cancer found in men worldwide, with around one in nine men being diagnosed with PCa within their lifetime. PCa often shows no symptoms in its early stages and its diagnosis techniques are either invasive, resource intensive, or has low efficacy, making widespread early detection onerous. Inspired by the recent success of deep convolutional neural networks (CNN) in computer aided detection (CADe), we propose a new CNN based framework for incidental detection of clinically significant prostate cancer (csPCa) in patients who had a CT scan of the abdomen/pelvis for other reasons. While CT is generally considered insufficient to diagnose PCa due to its inferior soft tissue characterisation, our evaluations on a relatively large dataset consisting of 139 clinically significant PCa patients and 432 controls show that the proposed deep neural network pipeline can detect csPCa patients at a level that is suitable for incidental detection. The proposed pipeline achieved an area under the receiver operating characteristic curve (ROC-AUC) of 0.88 (95% Confidence Interval: 0.86–0.90) at patient level csPCa detection on CT, significantly higher than the AUCs achieved by two radiologists (0.61 and 0.70) on the same task.

The task of learning from point cloud data is always challenging due to the often occurrence of noise and outliers in the data. Such data inaccuracies can significantly influence the performance of state-of-the-art deep learning networks and their ability to classify or segment objects. While there are some robust deep learning approaches, they are computationally too expensive for real-time applications. This paper proposes a deep learning solution that includes a novel robust pooling layer which greatly enhances network robustness and performs significantly faster than state-of-the-art approaches. The proposed pooling layer looks for data a mode/cluster using two methods, RANSAC, and histogram, as clusters are indicative of models. We tested the pooling layer into frameworks such as Point-based and graph-based neural networks, and the tests showed enhanced robustness as compared to robust state-of-the-art methods., Comment: under consideration at Computer Vision and Image Understanding

Robust model fitting is a fundamental problem in computer vision: used to pre-process raw data in the presence of outliers. Maximisation of Consensus (MaxCon) is one of the most popular robust criteria and widely used. Recently (Tennakoon et al. CVPR2021), a connection has been made between MaxCon and estimation of influences of a Monotone Boolean function. Equipping the Boolean cube with different measures and adopting different sampling strategies (two sides of the same coin) can have differing effects: which leads to the current study. This paper studies the concept of weighted influences for solving MaxCon. In particular, we study endowing the Boolean cube with the Bernoulli measure and performing biased (as opposed to uniform) sampling. Theoretically, we prove the weighted influences, under this measure, of points belonging to larger structures are smaller than those of points belonging to smaller structures in general. We also consider another "natural" family of sampling/weighting strategies, sampling with uniform measure concentrated on a particular (Hamming) level of the cube. Based on weighted sampling, we modify the algorithm of Tennakoon et al., and test on both synthetic and real datasets. This paper is not promoting a new approach per se, but rather studying the issue of weighted sampling. Accordingly, we are not claiming to have produced a superior algorithm: rather we show some modest gains of Bernoulli sampling, and we illuminate some of the interactions between structure in data and weighted sampling.

Low visibility hazard detection in construction sites is a crucial task for prevention of fatal accidents. Manual monitoring of construction workers to ensure they follow the safety rules (e.g., wear high-visibility vests) is a cumbersome task and practically infeasible in many applications. Therefore, an automated monitoring system is of both fundamental and practical interest. This paper proposes an intelligent solution that uses live camera images to detect workers who breach safety rules by not wearing high-visibility vests. The proposed solution is formulated in the form of an anomaly detection algorithm developed in the random finite set (RFS) framework. The proposed system is comprised of three steps: 1) applying a deep neural network to extract people in the image; 2) extracting particularly engineered features from each blob returned by the deep neural network; and 3) applying the RFS-based anomaly detection algorithm to each set of detected features. The experimental results demonstrate that in terms of F1-score, the proposed solution (as the combination of the newly engineered features and RFS-based anomaly detection algorithm) significantly outperforms various combinations of common and the state-of-the-art features and anomaly detection algorithms employed in machine vision applications.

Three-dimensional point clouds produced by 3D scanners are often noisy and contain outliers. Such data inaccuracies can significantly affect current deep learning-based methods and reduce their ability to classify objects. Most deep neural networks-based object classification methods were targeted to achieve high classification accuracy without considering classification robustness. Thus, despite their great success, they still fail to achieve good classification accuracy with low levels of noise and outliers. This work is carried out to develop a robust network structure that can solidly identify objects. The proposed method uses patches of planar segments, which can robustly capture object appearance. The planar segments information are then fed into a deep neural network for classification. We base our approach on the PointNet deep learning architecture. Our method was tested against several kinds of data inaccuracies such as scattered outliers, clustered outliers, noise and missing points. The proposed method shows excellent performance in the presence of these inaccuracies compared to state-of-the-art techniques. By decomposing objects into planes, the suggested method is simple, fast, provides good classification accuracy and can handle different kinds of point cloud data inaccuracies. The code can be found at https://github.com/AymanMukh/Pl-Net3D.

This paper presents an innovative method for motion segmentation in RGB-D dynamic videos with multiple moving objects. The focus is on finding static, small or slow moving objects (often overlooked by other methods) that their inclusion can improve the motion segmentation results. In our approach, semantic object based segmentation and motion cues are combined to estimate the number of moving objects, their motion parameters and perform segmentation. Selective object-based sampling and correspondence matching are used to estimate object specific motion parameters. The main issue with such an approach is the over segmentation of moving parts due to the fact that different objects can have the same motion (e.g. background objects). To resolve this issue, we propose to identify objects with similar motions by characterizing each motion by a distribution of a simple metric and using a statistical inference theory to assess their similarities. To demonstrate the significance of the proposed statistical inference, we present an ablation study, with and without static objects inclusion, on SLAM accuracy using the TUM-RGBD dataset. To test the effectiveness of the proposed method for finding small or slow moving objects, we applied the method to RGB-D MultiBody and SBM-RGBD motion segmentation datasets. The results showed that we can improve the accuracy of motion segmentation for small objects while remaining competitive on overall measures.

This paper approaches the problem of geometric multi-model fitting as a data segmentation problem. The proposed solution is based on a sequence of sampling hyperedges from a hypergraph, model selection and hypergraph clustering steps. We developed a sampling method that significantly facilitates solving the segmentation problem using a new form of the Markov-Chain-Monte-Carlo (MCMC) method to effectively sample from hyperedge distribution. To sample from this distribution effectively, our proposed Markov Chain includes new ways of long and short jumps to perform exploration and exploitation of all structures. To enhance the quality of samples, a greedy algorithm is used to exploit nearby structure based on the minimization of the Least kth Order Statistics cost function. Unlike common sampling methods, ours does not require any specific prior knowledge about the distribution of models. The output set of samples leads to a clustering solution by which the final model parameters for each segment are obtained. The method competes favorably with the state-of-the-art both in terms of computation power and segmentation accuracy.

Volumetric imaging is an essential diagnostic tool for medical practitioners. The use of popular techniques such as convolutional neural networks (CNN) for analysis of volumetric images is constrained by the availability of detailed (with local annotations) training data and GPU memory. In this paper, the volumetric image classification problem is posed as a multiinstance classification problem and a novel method is proposed to adaptively select positive instances from positive bags during the training phase. This method uses the extreme value theory to model the feature distribution of the images without a pathology and use it to identify positive instances of an imaged pathology. The experimental results, on three separate image classification tasks (i.e. classify retinal OCT images according to the presence or absence of fluid build-ups, emphysema detection in pulmonary 3D-CT images and detection of cancerous regions in 2D histopathology images) show that the proposed method produces classifiers that have similar performance to fully supervised methods and achieves the state of the art performance in all examined test cases.

The accuracy of handwritten text recognition may be affected by the presence of struck-out text in the handwritten manuscript. This paper investigates and improves the performance of a widely used handwritten text recognition approach Convolutional Recurrent Neural Network (CRNN) on handwritten lines containing struck out words. For this purpose, some common types of struck-out strokes were superimposed on words in a text line. A model, trained on the IAM line database was tested on lines containing struck-out words. The Character Error Rate (CER) increased from 0.09 to 0.11. This model was re-trained on dataset containing struck-out text. The model performed well in terms of struck-out text detection. We found that after providing an adequate number of training examples, the model can deal with learning struck-out patterns in a way that does not affect the overall recognition accuracy.

With the continuous process of urbanization in many cities around the globe, construction works are increasingly interrupting traffic flow in urban roads. At construction sites, usually traffic control personnel are employed to instruct the traffic flow at the times obstruction. Employing manual labour flagmen for traffic control is costly and can potentially expose the workers to safety hazards. Alternative attempts such as using remotely controlled signs or timers are not ideal; the former still requires human input and the latter does not optimize traffic control based on traffic density. To tackle this problem, we propose a fully automated traffic controller solution that uses visual sensors and machine vision technology. This system can replace the human traffic controllers during lane closures on the event of a construction. The implementation of this system eliminates the requirement for conventional traffic controllers, reducing the overall health and safety risks to workers and drivers. Our solution can result in substantial cost savings by replacing the traffic controllers. In addition, with manual labour, traffic control performance can be affected by weather conditions or fatigue from long hour shifts. Thus, our solution can lead to improved road safety and robustness of traffic flow as well.

Classification of 3D shapes into physically meaningful categories is one of the most important tasks in understanding the immediate environment. Methods that leverage the recent advancements in deep learning have shown to outperform the traditional approaches. However, performances of those methods have only been analyzed with relatively clean data. Three-dimensional measurement sets (point clouds) produced by 3D scanners are rarely that accurate and often contain noise, outliers or missing points. This paper presents an extensive analysis of the robustness of the state-of-the-art neural network algorithms to realistic data inaccuracies. Our experiments show that the existence of these inaccuracies can significantly affect the performance of the deep learning-based algorithms.

Predicting the presence of a disease in volumetric images is an essential task in medical imaging. The use of state-of-the-art techniques like deep convolutional neural networks (CNN) for such tasks is challenging due to limited supervised training data and high memory usage. This paper presents a weakly supervised solution that can be used in learning deep CNN features for volumetric image classification. In the proposed method, we use extreme value theory to model the feature distribution of the images without a pathology and use it to identify positive instances in an image that contains pathology. The experimental results show that the proposed method can learn classifiers that have similar performance to a fully supervised method and have significantly better performance in comparison with methods that use fixed number of instances from a positive image.

In this paper we present a novel non-Bayesian filtering method for tracking multiple objects with a particular application in time-lapse cell microscopic video sequence. In our method the heat-map of the frame sequence is extracted and represented as a pseudo-probability hypothesis density of the image. The pseudo-probability hypothesis density is used as measurements and fused with a prior Poisson random finite set density. We employed Cauchy-Schwarz divergence for information fusion. The presented algorithm was tested on a publicly available cell microscopic video sequence.

This paper proposes a novel method in order to obtain voxel-level segmentation for three fluid lesion types (IR-F/SRF/PED) in OCT images provided by the ReTOUCH challenge [1]. The method is based on a deep neural network consisting of encoding and de-coding blocks connected with skip-connections which was trained using a combined cost function comprising of cross-entropy, dice and adversarial loss terms. The segmentation results on a held-out validation set shows that the network architecture and the loss functions used has resulted in improved retinal fluid segmentation. Our method was ranked fourth in the ReTOUCH challenge.

Identifying the underlying models in a set of data points that is contaminated by noise and outliers leads to a highly complex multi-model fitting problem. This problem can be posed as a clustering problem by the projection of higher-order affinities between data points into a graph, which can be clustered using spectral clustering. Calculating all possible higher-order affinities is computationally expensive. Hence, in most cases, only a subset is used. In this paper, we propose an effective sampling method for obtaining a highly accurate approximation of the full graph, which is required to solve multi-structural model fitting problems in computer vision. The proposed method is based on the observation that the usefulness of a graph for segmentation improves as the distribution of the hypotheses that are used to build the graph approaches the distribution of the actual parameters for the given data. In this paper, we approximate this actual parameter distribution by using a $k$ th-order statistics-based cost function, and the samples are generated using a greedy algorithm that is coupled with a data sub-sampling strategy. The experimental analysis shows that the proposed method is both accurate and computationally efficient compared with the state-of-the-art robust multi-model fitting techniques. The implementation of the method is publicly available from https://github.com/RuwanT/model-fitting-cbs .

The fovea is one of the most important anatomical landmarks in the eye and its localization is required in automated analysis of retinal diseases due to its role in sharp central vision. In this paper, we propose a two-stage deep learning framework for accurate segmentation of the fovea in retinal colour fundus images. In the first stage, coarse segmentation is performed to localize the fovea in the fundus image. The location information from the first stage is then used to perform fine-grained segmentation of the fovea region in the second stage. The proposed method performs end-to-end pixelwise segmentation by creating a deep learning model based on fully convolutional neural networks, which does not require the prior knowledge of the location of other retinal structures such as optic disc (OD) and vasculature geometry. We demonstrate the effectiveness of our method on a dataset with 400 retinal images with average localization error of 14 ± 7 pixels.

Diabetic Retinopathy (DR) is one of the leading causes of blindness worldwide. Detecting DR and grading its severity is essential for disease treatment. Convolutional neural networks (CNNs) have achieved state-of-the-art performance in many different visual classification tasks. In this paper, we propose to combine CNNs with dictionary based approaches, which incorporates pathology specific image representation into the learning framework, for improved DR severity classification. Specifically, we construct discriminative and generative pathology histograms and combine them with feature representations extracted from fully connected CNN layers. Our experimental results indicate that the proposed method shows improvement in quadratic kappa score (κ2 = 0.86) compared to the state-of-the-art CNN based method (κ2 = 0.81).

This paper presents a new solution for statistical fusion of multi-sensor information acquired from different fields of view, in a centralized sensor network. The focus is on applications that involve tracking unknown number of objects with time-varying states. Our solution is a track-to-track fusion method in which the information contents of posteriors are combined. Existing information-theoretic solutions for track-to-track fusion in sensor networks are commonly devised based on minimizing the average information divergence from the local posteriors to the fused one. A common approach is to use Generalized Covariance Intersection rule for sensor fusion. This approach works best when all the sensors detect the same object(s), and performs poorly when fields-of-view are different. We suggest Cauchy–Schwarz divergence to be used for measuring information divergence. We demonstrate that employing Cauchy–Schwarz divergence leads to fusion rules that are generally more tolerant to imperfect consensus. We show that the proposed fusion rule for multiple Poisson posteriors is the weighted arithmetic mean of the Poisson densities. Furthermore, we derive the fusion rule for labeled multi Bernoulli filter by approximating the labeled multi Bernoulli density to its first order moment. Numerical experiments show the superior performance of our solution compared to Kullback–Leibler averaging method.

Complexities of dynamic volumetric imaging challenge the available computer vision techniques on a number of different fronts. This paper examines the relationship between the estimation accuracy and required amount of smoothness for a general solution from a robust statistics perspective. We show that a (surprisingly) small amount of local smoothing is required to satisfy both the necessary and sufficient conditions for accurate optic flow estimation. This notion is called 'just enough' smoothing, and its proper implementation has a profound effect on the preservation of local information in processing 3D dynamic scans. To demonstrate the effect of 'just enough' smoothing, a robust 3D optic flow method with quantized local smoothing is presented, and the effect of local smoothing on the accuracy of motion estimation in dynamic lung CT images is examined using both synthetic and real image sequences with ground truth.

This paper approaches the problem of geometric multi-model fitting as a data segmentation problem which is solved by a sequence of sampling, model selection and clustering steps. We propose a sampling method that significantly facilitates solving the segmentation problem using the Normalized cut. The sampler is a novel application of Markov-Chain-Monte-Carlo (MCMC) method to sample from a distribution in the parameter space that is obtained by modifying the Least kth Order Statistics cost function. To sample from this distribution effectively, our proposed Markov Chain includes novel long and short jumps to ensure exploration and exploitation of all structures. It also includes fast local optimization steps to target all, even fairly small, putative structures. This leads to a clustering solution through which final model parameters for each segment are obtained. The method competes favorably with the state-of-the-art both in terms of computation power and segmentation accuracy.

The quality of input images significantly affects the outcome of automated diabetic retinopathy screening systems. Current methods to identify image quality rely on hand-crafted geometric and structural features, that does not generalize well. We propose a new method for retinal image quality classification (IQC) that uses computational algorithms imitating the working of the human visual systems. The proposed method leverages on learned supervised information using convolutional neural networks (CNN), thus avoiding hand-engineered features. Our analysis shows that the learned features capture both geometric and structural information relevant for image quality classification. Experimental results conducted on a relatively large dataset demonstrates that the overall method can achieve high accuracy. We also show that effective features for IQC can be learned by full training of shallow CNN as well as by using transfer learning.

This paper presents a track-before-detect labeled multi-Bernoulli filter tailored for industrial mobile platform safety applications. We derive two application specific separable likelihood functions that capture the geometric shape and colour information of the human targets who are wearing a high visible vest. These likelihoods are then used in a labeled multi-Bernoulli filter with a novel two step Bayesian update. Preliminary simulation results show that the proposed solution can successfully track human workers wearing a luminous yellow colour vest in an industrial environment., The conference which this paper was submitted, has rejected this paper. Thus, we are in the process of enhancing the content of the paper and submit it to another conference/journal

Image registration is the task of finding a spatial transformation T that aligns the objects in two or more images capturing the same or related scene. It is one of the most crucial problems of computer vision and has been studied for over three decades. The underlying task is very general and has a wide range of applications in the fields of medical imaging, machine vision, remote sensing, cartography, etc. In this chapter we focus on the application of image registration in the fields of medical imaging.

Using splines to model spatio-temporal data is one of the most common methods of data fitting used in a variety of computer vision applications. Despite its ubiquitous applications, particularly for volumetric image registration and interpolation, the existing estimation methods are still sensitive to the existence of noise and outliers. A method of robust data modelling using thin plate splines, based upon the well-known least K-th order statistical model fitting, is proposed and compared with the best available robust spline fitting techniques. Our experiments show that existing methods are not suitable for typical computer vision applications where outliers are structured (pseudo-outliers) while the proposed method performs well even when there are numerous pseudo-outliers.

Nonrigid image registration techniques using intensity based similarity measures are widely used in medical imaging applications. Due to high computational complexities of these techniques, particularly for volumetric images, finding appropriate registration methods to both reduce the computation burden and increase the registration accuracy has become an intensive area of research. In this paper, we propose a fast and accurate nonrigid registration method for intra-modality volumetric images. Our approach exploits the information provided by an order statistics based segmentation method, to find the important regions for registration and use an appropriate sampling scheme to target those areas and reduce the registration computation time. A unique advantage of the proposed method is its ability to identify the point of diminishing returns and stop the registration process. Our experiments on registration of end-inhale to end-exhale lung CT scan pairs, with expert annotated landmarks, show that the new method is both faster and more accurate than the state of the art sampling based techniques, particularly for registration of images with large deformations.

Condition monitoring of storm-water pipe systems are carried-out regularly using semi-automated processors. Semi-automated inspection is time consuming, expensive and produces varying and relatively unreliable results due to operators fatigue and novicity. This paper propose an innovative method to automate the storm-water pipe inspection and condition assessment process which employs a computer vision algorithm based on deep-neural network architecture to classify the defect types automatically. With the proposed method, the operator only needs to guide the robot through each pipe and no longer needs to be an expert. The results obtained on a CCTV video dataset of storm-water pipes shows that the deep neural network architectures trained with data augmentation and transfer learning is capable of achieving high accuracies in identifying the defect types.