In this paper, a multi-level vehicle crash detection is proposed. The first step is to detect and track the movement of the vehicles in a scene and based on the stop condition of any vehicle, we trigger the first leve...
详细信息
ISBN:
(纸本)9781538643716
In this paper, a multi-level vehicle crash detection is proposed. The first step is to detect and track the movement of the vehicles in a scene and based on the stop condition of any vehicle, we trigger the first level which is a possible crash detected. Next we scan the object with a pre-trained cascade detector looking for possible defects in it. The areas detected are passed into a verification function which is based on a pre-trained SVM to classify their HOG if they represent area of damage in a vehicle or not. We tested static images of few cars and we were able to detect and verify the defect in their body. Then we built a webcam simulator and a web application to test the solution in a real-time environment and over the web. The test succeeded and we were able to detect a real accident successfully.
Widely sharing image data captured by personal or surveillance cameras can enable a variety of research studies, e.g., computervision, intelligent systems, and social sciences, to name a few. However, image data may ...
详细信息
ISBN:
(纸本)9781538692141
Widely sharing image data captured by personal or surveillance cameras can enable a variety of research studies, e.g., computervision, intelligent systems, and social sciences, to name a few. However, image data may contain a range of sensitive information, such as identity, location, and health, which creates significant challenges for sharing image data with untrusted parties, e.g., researchers. Standard obfuscation methods obscure regions-of-interest (ROIs) in image data with pixelization and blurring, which do not offer formal privacy guarantees. Moreover, such obfuscated image data can be re-identified by convolutional neural networks. In this demonstration, we will showcase a novel method for sanitizing sensitive image ROIs with quantifiable privacy guarantees. The audience will observe the obfuscation in a camera live stream, which demonstrates the feasibility of our method for real-time applications. Furthermore, the audience can interact with our method by selecting photos from publicly available datasets and choosing different privacy levels.
In this paper are presented some hardware and software solutions for the blind people. Only in Bucharest, the capital of Romania, there are over 6000 blind people, and the number of people with acute vision problems i...
详细信息
ISBN:
(纸本)9781728107738
In this paper are presented some hardware and software solutions for the blind people. Only in Bucharest, the capital of Romania, there are over 6000 blind people, and the number of people with acute vision problems is much higher. There are software applications designed to help these people in their daily lives. With today's advanced technology, there are smartphones, tablets, or personal computer (PC) smartphone applications that help the blind persons for traveling and reading information. The software applications proposed in this article are useful for people with visual impairments in order to learn how to use a PC keyboard. The hardware device proposed can be a solution for learning to use a 3x4 numeric keypad by blind person.
Transformers have improved drastically the performance of natural language processing (NLP) and computervisionapplications. The computation of transformers involves matrix multiplications and non-linear activation f...
详细信息
ISBN:
(纸本)9798350383638;9798350383645
Transformers have improved drastically the performance of natural language processing (NLP) and computervisionapplications. The computation of transformers involves matrix multiplications and non-linear activation functions such as softmax and GELU (Gaussion Error Linear Unit) that are accelerated directly in hardware. Currently, function evaluation is done separately for each function and rarely allows for hardware reuse. To mitigate this problem, in this work, we map the computation of GELU to a softmax operator. In this way, the efficient hardware units designed already for softmax can be reused for computing GELU as well. Computation of GELU can enjoy the inherent vectorized nature of softmax and produce in parallel multiple GELU outcomes. Experimental results show that computing GELU via a pre-existing and incrementally modified softmax hardware unit (a) does not reduce the accuracy of representative NLP applications and (b) allows the reduction of the overall hardware area and power by 6.1% and 11.9%, respectively, on average.
Recent modeling experiments conducted in computational music give evidence that a number of concepts, methods and tools belonging to inverse semigroup theory can be attuned towards the concrete modeling of time-sensit...
详细信息
ISBN:
(纸本)9783319042978;9783319042985
Recent modeling experiments conducted in computational music give evidence that a number of concepts, methods and tools belonging to inverse semigroup theory can be attuned towards the concrete modeling of time-sensitive interactive systems. Further theoretical developments show that some related notions of higher-dimensional strings can be used as a unifying theme across word or tree automata theory. In this invited paper, we will provide a guided tour of this emerging theory both as an abstract theory and with a view to concrete applications.
Concepts from information theory have recently found favor in both the mainstream computervision community and the military automatic target recognition community. In the computervision literature, the principles of...
详细信息
ISBN:
(纸本)0819440744
Concepts from information theory have recently found favor in both the mainstream computervision community and the military automatic target recognition community. In the computervision literature, the principles of minimax entropy learning theory have been used to generate rich probabilitistic models of texture and shape. In addition, the method of types and large deviation theory has permitted the difficulty of various texture and shape recognition tasks to be characterized by "order parameters" that. determine how fundamentally vexing a task is, independent of the particular algorithm used. These information-theoretic techniques have been demonstrated using traditional visual imagery in applications such as simulating cheetah skin textures and such as finding roads in aerial imagery. We discuss their application to problems in the specific application domain of automatic target recognition using infrared imagery. We also review recent theoretical and algorithmic developments which permit learning minimax entropy texture models for infrared textures in reasonable timeframes.
Palynology, the study of pollen, is becoming the focus of attention in computervision in recent years. Various proposed automated classification and segmentation methods have been evaluated on a number of data sets. ...
详细信息
ISBN:
(纸本)9781665443371
Palynology, the study of pollen, is becoming the focus of attention in computervision in recent years. Various proposed automated classification and segmentation methods have been evaluated on a number of data sets. However, as of 2021 most data sets are sparse;they either contain only a small number of pollen classes, images in total or are imbalanced overall. In this work, we explore the possibility of creating synthetic pollen grain images from less than 2,000 images per pollen class via a Generative Adversarial Network (GAN). For that purpose, we selected two distinct pollen classes from a state of the art pollen data set and evaluated the data set with and without synthetic data on a Convolutional Neural Network (CNN). The enriched data set performed better overall (+1.4%) and specifically for the two pollen classes (+2%). We also drastically reduced the no. of real images and were still able to achieve a score of 60% to 80%. The experiments show, that our synthesized pollen images are visually close to real-life pollen grains and can be used to enrich imbalanced data sets as an addition to traditional data augmentation methods.
This paper presents an innovative approach to solving complex multi-objective optimization problems through an asynchronous and distributed evolutionary game theory method. The proposed algorithm, an extension of the ...
详细信息
ISBN:
(纸本)9798331531317;9798331531300
This paper presents an innovative approach to solving complex multi-objective optimization problems through an asynchronous and distributed evolutionary game theory method. The proposed algorithm, an extension of the IMGAMO algorithm, optimizes individual criteria separately at varying computational speeds, thus significantly enhancing computational efficiency and adaptability. This unique structure enables independent criterion optimization, catering to real-world applications where different objectives demand varying computational resources. The algorithm's effectiveness is validated against traditional synchronous evolutionary multi-objective optimization algorithms, showing superior performance in handling diverse, real-world problems efficiently. The results underline the potential of the asynchronous approach in providing high-quality Pareto fronts, thus offering robust solutions for complex optimization challenges.
Homography estimation is an important image alignment method that has been widely used in computervisionapplications. Traditional methods heavily rely on the distribution of features and usually fail in low-texture ...
详细信息
ISBN:
(纸本)9781665468916
Homography estimation is an important image alignment method that has been widely used in computervisionapplications. Traditional methods heavily rely on the distribution of features and usually fail in low-texture and large-baseline scenes. Most learning-based methods use convolutional neural networks(CNNs) to extract features. However, the dense features extracted in this way have a limited receptive field, leading to poor accuracy of results. In this paper, we propose a novel method for homography estimation. We first estimate the projective transformation between the reference image and the target image at a coarse level and then refine the estimated homography at the fine level. Unlike approaches that use simple CNNs or global correlations to search correspondences, we add self- and cross-attention layers in the transformer to enhance the feature correlations. The experiments show that our method significantly outperforms the existing solutions in challenging large-baseline scenes.
Blood cell counting is an important medical test for successful diagnosis of many diseases. Traditional blood cell counting tasks are tedious, time-consuming, and depend a lot on the skill of the equipment operator. T...
详细信息
ISBN:
(纸本)9798350361513;9798350372304
Blood cell counting is an important medical test for successful diagnosis of many diseases. Traditional blood cell counting tasks are tedious, time-consuming, and depend a lot on the skill of the equipment operator. Therefore, there is a need to automate this process and deep learning can be an effective solution. In this paper, two deep learning models are deployed to detect RBCs, WBCs, and Platelets in blood smear images. The You Only Look Once (YOLO) and EfficientDet models were trained on annotated blood smear images from three different publicly available datasets and the performance of the models is evaluated and compared. These trained models were also able to detect all the three types of blood cells on random blood smear images from other datasets, proving that they are generalized. Next, mobile applications were created in order to perform cell detection in real time. The results show that the process of blood cell count can be automated and this process can be much faster and more efficient in the future with the help of computer-Aided Diagnosis (CAD) models.
暂无评论