A Novel Classification Model for Suspicious Human Activities in Diverse Environments Using Fused Feature Block and Machine Vision Techniques

Suspicious Activity is predicting the body part of a person from video. This project will entail detecting suspicious human Activity from video using neural networks. Suspicious human activity detection from surveillance video is an active research area of image processing and computer vision. Using visual surveillance, human activities can be monitored in public areas such as bus stations, railway stations, airports, banks, shopping malls, school and colleges, parking lots, roads, etc. to prevent terrorism, accidents and illegal parking, vandalism, fighting, crime and other suspicious activities. It is very tough to watch public places continuously, so we use an intelligent video surveillance, It is required to monitor the human activities from video and categorize them as usual and unusual activities; and can generate an alert.

Abstract: Suspicious human activity detection in security capture is a study topic in image processing and vision. The mysterious identification of human activity from video surveillance is an area of study in both fields. Human activity can be monitored visually in conspicuous public spaces like bus depots, airports, railway stations, financial institutions, malls, schools, and universities to avoid terrorist activity, vandalism, accidents, prohibited parking spaces, vandalism, fighting chain theft, criminality, and other unusual behavior. Extremely difficult to continually monitor public spaces, thereby an innovative video surveillance installation system that can track people's movements in real-time, classify them as routine or odd, and send out an alert is needed. The field of visual surveillance to identify aberrant actions has seen a significant amount of publications in the last ten years. Furthermore There are a few surveys in the literature for recognising various abnormal activities, but none have reviewed various abnormal activities, but none of them have reviewed various abnormal activities. This study presents the stateof-the-art in the field of recognizing suspicious behavior from surveillance recordings during the past tenyears. We provide a brief outline of the risks and challenges associated with detecting suspicious human activity. This article examines six aberrant behaviors, including the identification of abandoned objects, theft, falls, traffic accidents, and unlawful parking, as wellas the detection of violence and fire. Generally speaking, we have covered all the processes that have been [1] Foreground object extraction, object identification based on tracking or non-tracking approaches, feature extraction, classification, activity analysis, and recognition are some ofthe techniques that have been used to identify human activity from surveillance movies in the literature. This paper's goal is to give field researchers a literature assessment of six different suspicious activity identificationsystems together with its broad framework.[1]

Abstract—With the unprecedented development of technolo- gies, there has been the trend of rapid deployment of surveillance systems in multiple domains, such as security and healthcare, and public safety. In this paper, a new method of suspicious human activity detection based on Convolutional Neural Net- works (CNNs) combined with Long Short-Term Memory (LSTM) networks is proposed. The presented system seeks to improve the robustness and speed of anomaly detection across real-time video streams. Leveraging CNNs for spatial feature extraction and LSTMs for temporal sequence modeling, the proposed system efficiently detects suspicious activities in real time. Index Terms—Suspicious Activity Detection, Human Activity Recognition, Convolutional Neural Networks, Deep Learning, Video Surveillance

Due to the fact that there are problems in neurophysiological research on post-traumatic epilepsy in order to find sleep spindles and epileptiform discharges in long-term (day or more) recordings of electroencephalography (EEG) there is a need for algorithms for automatic detection of suspicious EEG activity (we call as suspicious activity any EEG activity that differs from the background activity). There are many methods of signal processing. The most common and straightforward are the methods of transition from the temporal representation of the signal to the time-frequency representation. One of them is the wavelet transform. For the wavelet spectrograms, the ridges of the wavelet spectrograms are calculated. Method of detecting the suspicious activity involves an analysis of points of the ridges. The spectrogram of ridge points are calculated, after which the points of the ridge are divided into two groups: those that relate to the background activities and those that relate to suspicious activity. Suspicious activity that does not meet the requirements of neuroscientists is eliminated.

Detection of suspicious human actions in automated video surveillance applications, is of great practical importance. Those kind of unusual activities in human is very difficult to acquire and classify to predict. In our proposed work, automatic tracking and detecting unusual movement’s problems in closed circuit videos was resolved. Firstly, the videos are converted into frames. Then from the obtained frames, humans are detected from the video using a background subtraction method. Then the features are extracted using a convolutional neural network (CNN). The features thus extracted are fed to a Discriminative Deep Belief Network (DDBN). Labeled videos of some suspicious activities are also fed to the DDBN and their features are also extracted. Then the features extracted using Convolutional Neural Network (CNN) are compared against these features extracted from the labeled sample video of classified suspicious actions using a Discriminative Deep Belief Network (DDBN) and various suspicious activities are detected from the given video and results shows increase accuracy of 90% for the proposed framework for classification.

Suspicious human activity detection is a major area of research and development that focuses on sophisticated machine learning techniques to reduce monitoring costs while enhancing safety. Since it is difficult for people to continually monitor public spaces, we need a real-time intelligent human activity recognition system that can identify suspicious activities. Current systems use low-accurate complex algorithms and techniques, making the system less reliable. This paper proposes a real-time suspicious human activity recognition with high accuracy by introducing a Convolutional Neural Network and using the 2D pose estimation technique to the system. This system can be used for home security, hospitals, and other areas of surveillance. Here, we are extracting skeletal images of humans from the input video frames using 2D pose estimation to identify the pose of humans in the videos. These poses are then passed to a pre-trained Convolutional Neural Network to classify different activities of humans like trespassing or not trespassing, fall or not fall, fighting, etc. After analyzing the pixels and activities, an alert can be produced through alarms, messages to phones, email the footage to the owner or security professional, and other techniques to prevent unusual activities. This system can be used in public places like shopping malls, railway stations, public roads, and even in homes, universities, and educational institutions.

Human activity detection using video surveillance system is the processing of consecutive video frames and analysing any suspicious activity occurring in the video. Detection of human activities comes under the areas of artificial intelligence with image analysis and computer vision as sub-domain. The paper deals with detection of harmful or troublesome human activities using the matching of features obtained by SIFT, so that the further suspicious action can be prevented by alerting the concerned system. The paper narrates about the detection of suspicious activity such as holding a gun, wielding a knife, or punching based on their features. The proposed approach analyses the video frame by frame by observing the suspicious object as well as its activities in the video, and implements SIFT for extraction of features and mean shift algorithm (MSA) for object tracking. This paper gives an apparent idea of developing the system detecting suspicious activities of humans.

A new approach based on a 1D + 2D convolutional neural network and evolving fuzzy system for the diagnosis of cardiovascular disease from heart sound signals

The content of tea polyphenols (TP) is one of the important indicators for judging the quality of tea. Accurate and non-destructive estimation technology for tea polyphenol content has attracted more and more attention, which has become a key technology for tea production, quality identification, grading and so on. Hyperspectral imaging technology is a fusion of spectral analysis and image processing technology, which has been proven to be an efficient technology for predicting tea polyphenol content. To make full use of spectral and spatial features, a prediction model of tea polyphenols based on spectral-spatial deep features extracted using convolutional neural network (CNN) was proposed, which not only broke the limitations of traditional shallow features, but also innovated the technical path of integrated deep learning in non-destructive detection for tea. Firstly, one-dimensional convolutional neural network (1D-CNN) and two-dimensional convolutional neural network (2D-CNN) models were constructed to extract the spectral deep features and spatial deep features of tea hyperspectral images, respectively. Secondly, spectral deep features, spatial deep features, and spectral-spatial deep features are used as input variables of machine learning models, including Partial Least Squares Regression (PLSR), Support Vector Regression (SVR) and Random Forest (RF). Finally, the training, testing and evaluation were realized using the self-built hyperspectral dataset of green tea from different grades and different manufacturers. The results showed that the model based on spectral-spatial deep features had the best prediction performance among the three machine learning models (R2 = 0.949, MAE = 0.533 for training sets, R2 = 0.938, MAE = 0.799 for test sets). Moreover, the visualization of estimation results of tea polyphenol content further demonstrated that the model proposed in this study had strong estimation ability. Therefore, the deep features extracted using CNN can provide new ideas for estimation of the main components of tea, which will provide technical support for the estimation tea quality estimation.

Preoperative Knowledge of the histological grade of hepatocellular carcinoma (HCC) is significant for patient management and prognosis in clinical practice. Recent studies reported that 3D Convolutional Neural Network (CNN) outperformed 2D CNN for lesion characterization. Since 2D and 3D deep feature derived from CNN embed different spatial information of neoplasm, we hypothesize that the performance of lesion characterization might be improved if taking full advantage of both 2D and 3D characterization. In this work, we propose a 2D and 3D CNN fusion architecture to integrate both 2D and 3D spatial information of neoplasm for predicting the histological grade of HCC. Specifically, correlated and individual component analysis (CICA) is performed to fuse the 2D deep features in three orthogonal views and the 3D deep feature in volumetric images of HCC. Experimental results of 46 clinical patients with HCCs demonstrate several encouraging features of the proposed 2D and 3D deep feature fusion framework as follows: (1) Fusion of 2D and 3D deep feature using CICA outperforms 2D or 3D deep feature for predicting the histological grade of HCC. (2) Fusion of 2D deep features derived from three orthogonal views using CICA yields better results than those of 3D deep feature. (3) CICA is better than the conventional concatenation and the correlation learning model for deep feature fusion.

The detection of human beings in a camera attracts more attention because of its wide range of applications such as abnormal event detection, person counting in a dense crowd, person identification, fall detection for care to elderly people, etc. Over the time, various techniques have evolved to enhance the visual information. This article presents a novel 3-D intelligent information system for identifying abnormal human activity using background subtraction, rectification, morphology, neural networks and depth estimation with a thermal camera and a pair of hand held Universal Serial Bus (USB) camera to visualize un-calibrated images. The proposed system detects strongest points using Speed-Up Robust Features (SURF). The Sum of Absolute Difference (SAD) algorithm match the strongest points detected by SURF. 3-D object model and image stitching from image sequences are carried out in the proposed work. A series of images captured from different cameras are stitched into a geometrically consistent mosaic either horizontally/vertically based on the image acquisition. 3-D image and depth estimation of un-calibrated stereo images are acquired using rectification and disparity. The background is separated from the scene using threshold approach. Features are extracted using morphological operators in order to get the skeleton. Junction points and end points of the skeleton image are obtained from the skeleton. Data set of abnormal human activity is created using supervised learning such as neural network with a thermal camera and a pair of webcam. The feature vector of an activity is compared with already created data set, if a match occurs the classifier detects abnormal human activity. Additionally the proposed algorithm performs depth estimation to measure real time distance of objects dynamically. The system use thermal camera, Intel computing stick, converter, video graphics array (VGA) to high-definition multimedia interface (HDMI) and webcams. The proposed novel intelligent information system gives 94% maximum accuracy and 89% minimum accuracy for different activities, thus it effectively detects suspicious activity during day and night.

ABSTRACTPoint cloud classification, which provides meaningful semantic labels to the points in a point cloud, is essential for generating three-dimensional (3D) models. Its automation, however, remains challenging due to varying point densities and irregular point distributions. Adapting existing deep-learning approaches for two-dimensional (2D) image classification to point cloud classification is inefficient and results in the loss of information valuable for point cloud classification. In this article, a new approach that classifies point cloud directly in 3D is proposed. The approach uses multi-scale features generated by deep learning. It comprises three steps: (1) extract single-scale deep features using 3D convolutional neural network (CNN); (2) subsample the input point cloud at multiple scales, with the point cloud at each scale being an input to the 3D CNN, and combine deep features at multiple scales to form multi-scale and hierarchical features; and (3) retrieve the probabilities that each point belongs to the intended semantic category using a softmax regression classifier. The proposed approach was tested against two publicly available point cloud datasets to demonstrate its performance and compared to the results produced by other existing approaches. The experiment results achieved 96.89% overall accuracy on the Oakland dataset and 91.89% overall accuracy on the Europe dataset, which are the highest among the considered methods.

To effectively grade hepatocellular carcinoma (HCC) based on deep features derived from diffusion weighted images (DWI) with multiple b-values using convolutional neural networks (CNN). Ninety-eight subjects with 100 pathologically confirmed HCC lesions from July 2012 to October 2018 were included in this retrospective study, including 47 low-grade and 53 high-grade HCCs. DWI was performed for each subject with a 3.0T MR scanner in a breath-hold routine with three b-values (0,100, and 600s/mm2 ). First, logarithmic transformation was performed on original DWI images to generate log maps (logb0, logb100, and logb600). Then, a resampling method was performed to extract multiple 2D axial planes of HCCs from the log map to increase the dataset for training. Subsequently, 2D CNN was used to extract deep features of the log map for HCCs. Finally, fusion of deep features derived from three b-value log maps was conducted for HCC malignancy classification. Specifically, a deeply supervised loss function was devised to further improve the performance of lesion characterization. The data set was split into two parts: the training and validation set (60 HCCs) and the fixed test set (40 HCCs). Four-fold cross validation with 10 repetitions was performed to assess the performance of deep features extracted from single b-value images for HCC grading using the training and validation set. Receiver operating characteristic curve (ROC) and area under the curve (AUC) values were used to assess the characterization performance of the proposed deep feature fusion method to differentiate low-grade and high-grade in the fixed test set. The proposed fusion of deep features derived from logb0, logb100, and logb600 with deeply supervised loss function generated the highest accuracy for HCC grading (80%), thus outperforming the method of deep feature derived from the ADC map directly (72.5%), the original b0 (65%), b100 (68%), and b600 (70%) images. Furthermore, AUC values of the deep features of the ADC map, the deep feature fusion with concatenation, and the proposed deep feature fusion with deeply supervised loss function were 0.73, 0.78, and 0.83, respectively. The proposed fusion of deep features derived from the logarithm of the three b-value images yields high performance for HCC grading, thus providing a promising approach for the assessment of DWI in lesion characterization.

This paper presents a new algorithm for suspicious human activity recognition in videos based on a combination of two different feature types. The first feature concerns the shape and is called shape moments. The second concerns the boundary coordinates and is called Histogram of Normalized Distances (HND) from Center of gravity of the object shape (COG) and it's contour points combining these features leads to the formation of a strong complementary feature vector that captures effective discriminate details of human action videos. The authors used two methods for classification, the Multi-class Support Vector Machine and Naive Bayes classifier. The classification by using the Multi-class SVM classifier verified recognition rate up to 95.6 %, but the Naive Bayes classifier verified 97.2%. The authors evaluated the suspicious activity recognition on 250 videos from HMDB data set. Five distinct suspicious human activities (e.g., Running, Punching, Kicking, Shooting guns and Falling floor, etc.) by 250 different persons. Experiments on HMDB show that the presented system can recognize suspicious activities effectively and accurately in surveillance videos.

The acquisition of high-quality labeled synthetic aperture radar (SAR) data is challenging due to the demanding requirement for expert knowledge. Consequently, the presence of unreliable noisy labels is unavoidable, which results in performance degradation of SAR automatic target recognition (ATR). Existing research on learning with noisy labels mainly focuses on image data. However, the non-intuitive visual characteristics of SAR data are insufficient to achieve noise-robust learning. To address this problem, we propose collaborative learning of scattering and deep features (CLSDF) for SAR ATR with noisy labels. Specifically, a multi-model feature fusion framework is designed to integrate scattering and deep features. The attributed scattering centers (ASCs) are treated as dynamic graph structure data, and the extracted physical characteristics effectively enrich the representation of deep image features. Then, the samples with clean and noisy labels are divided by modeling the loss distribution with multiple class-wise Gaussian Mixture Models (GMMs). Afterward, the semi-supervised learning of two divergent branches is conducted based on the data divided by each other. Moreover, a joint distribution alignment strategy is introduced to enhance the reliability of co-guessed labels. Extensive experiments have been done on the Moving and Stationary Target Acquisition and Recognition (MSTAR) and SAR-ACD datasets, and the results show that the proposed method can achieve state-of-the-art performance under different operating conditions with various label noises. The code is released at https://github.com/fuyimin96/CLSDF.