Human Blobs Research Articles

Performance Evaluation of Neural Network for Human Classification Using Blob Dataset

Background: Human Classification in public places is an emerging area in the applications of Computational Intelligence. Therefore, modeling of an optimal architecture of the neural network is required to classify them. Methods: In this work for this purpose, blob dataset has been used to train the neural network. This dataset consists of 2408 features of a human blob. Results: Further, analysis of this blob dataset has been done on the basis of various characteristic parameters for affirmation of actual training. During training and testing of this dataset, it has been observed that when nodes at hidden layer are below and above 10 then training of neural network is under fitted and overfitted respectively and works effectively when the nodes are 10 at the hidden layer. Conclusion: From the experimental work performed in this study, an optimal neural network has been obtained to classify human using blob dataset.

Image Analysis Using Human Body Geometry and Size Proportion Science for Action Classification

Gestures are one of the basic modes of human communication and are usually used to represent different actions. Automatic recognition of these actions forms the basis for solving more complex problems like human behavior analysis, video surveillance, event detection, and sign language recognition, etc. Action recognition from images is a challenging task as the key information like temporal data, object trajectory, and optical flow are not available in still images. While measuring the size of different regions of the human body i.e., step size, arms span, length of the arm, forearm, and hand, etc., provides valuable clues for identification of the human actions. In this article, a framework for classification of the human actions is presented where humans are detected and localized through faster region-convolutional neural networks followed by morphological image processing techniques. Furthermore, geometric features from human blob are extracted and incorporated into the classification rules for the six human actions i.e., standing, walking, single-hand side wave, single-hand top wave, both hands side wave, and both hands top wave. The performance of the proposed technique has been evaluated using precision, recall, omission error, and commission error. The proposed technique has been comparatively analyzed in terms of overall accuracy with existing approaches showing that it performs well in contrast to its counterparts.

Automatic human trajectory destination prediction from video

This paper presents an intelligent human trajectory destination detection system from video. The system assumes a passive collection of video from a wide scene used by humans in their daily motion activities such as walking towards a door. The proposed system includes three main modules, namely human blob detection, star skeleton detection and destination area prediction, and it works directly with raw video, producing motion features for destination prediction system, such as position, velocity and acceleration from detected human skeletons, resulting in several input features that are used to train a machine learning classifier. We adopted a university campus exterior scene for the experimental study, which includes 348 pedestrian trajectories from 171 videos and five destination areas: A, B, C, D and E. A total of six data processing combinations and four machine learning classifiers were compared, under a realistic growing window evaluation. Overall, high quality results were achieved by the best model, which uses 37 skeleton motion inputs, undersampling on training data and a random forest. The global discrimination, in terms of area of the receiver operating characteristic curve is around 87%. Furthermore, the best model can predict in advance the five destination classes, obtaining a very good ahead discrimination for classes A, B, C and D, and a reasonable ahead discrimination for class E.

Recognizing human activities using appearance metric feature and kinematics feature

The problem of automatically recognizing human activities from videos through the fusion of the two most important cues, appearance metric feature and kinematics feature, is considered. And a system of two-dimensional (2-D) Poisson equations is introduced to extract the more discriminative appearance metric feature. Specifically, the moving human blobs are first detected out from the video by background subtraction technique to form a binary image sequence, from which the appearance feature designated as the motion accumulation image and the kinematics feature termed as centroid instantaneous velocity are extracted. Second, 2-D discrete Poisson equations are employed to reinterpret the motion accumulation image to produce a more differentiated Poisson silhouette image, from which the appearance feature vector is created through the dimension reduction technique called bidirectional 2-D principal component analysis, considering the balance between classification accuracy and time consumption. Finally, a cascaded classifier based on the nearest neighbor classifier and two directed acyclic graph support vector machine classifiers, integrated with the fusion of the appearance feature vector and centroid instantaneous velocity vector, is applied to recognize the human activities. Experimental results on the open databases and a homemade one confirm the recognition performance of the proposed algorithm.

Fast medium-scale multiperson identification in aerial videos

Vision systems for unmanned aerial vehicles (UAVs) have been gaining increasing attention for surveillance and civil applications. However, aerial platforms create new challenges for several vision tasks (e.g., human tracking and identification) because UAV-mounted cameras undergo large vibration movements and capture unstable videos. Furthermore, most existing machine vision approaches use the fine details of a human figure, which are unavailable in low-quality aerial images. We propose a new blob-matching approach for human identification in aerial videos in which the identity of a human blob is estimated using an adaptive reference set of previously identified people. A target can be quickly located by matching only the target and a carefully selected candidate set. The experimental results obtained using several challenging aerial videos validated the effectiveness and computational efficiency of the proposed method.

Recognition of human activities using a multiclass relevance vector machine

We address the issue of human activity recognition by introducing the multiclass relevance vector machine (mRVM), the current state-of-the-art kernel machine learning technology given the multiclass classification problems (actually, activity recognition can commonly be viewed as a multiclass classification problem). Under our proposed recognition framework, the required procedure consists of three functional cascade modules: a. detecting the human silhouette blobs from the image sequence by the background subtraction method; b. extracting the shape and the motion features from the variation energy image (VEI); and c. sending the obtained features to the mRVM and recognizing the human activity. There are two types of mRVM: the constructive mRVM1 and the top-down mRVM2. We performed 10 times three-fold cross-validation on the Weizmann benchmark data set to examine the effectiveness of the proposed method. We also compared our method with other existing approaches, and the experimental results show that the proposed method offers superior performance. In summary, the mRVM, especially the mRVM2, has advantages both in terms of recognition rate and sparsity, along with a simple feature extraction process. The mRVM also significantly simplifies the classification process, by comparison with traditional binary-tree style multiclass classifiers.

Detection and matching of multiple occluded moving people for human tracking in colour video sequences

The proposed approach aims to track multiple moving people in a colour video acquired with a single camera. The first phase of the approach consists in precisely detecting multi-human inside moving foregrounds. The input to this phase is foreground pixels which were extracted from the scene using any background subtraction technique. These moving foregrounds are then further segmented into multiple moving people using region segmentation and shape-based occlusion handling. The second phase assigns the detected human blobs to tracks using robust matching process based both on appearance model and motion model. For this, we use Kalman filter to predict future locations and sizes for dynamic persons and fuse this information with appearance-based comparison in order to assign each blob to a track. The preliminary experiments on several representative sequences have shown that this unsupervised approach can robustly detect and track multiple occluded moving persons, even at lower temporal resolution.

Recognition of human activities using SVM multi-class classifier

Even great efforts have been made for decades, the recognition of human activities is still an unmature technology that attracted plenty of people in computer vision. In this paper, a system framework is presented to recognize multiple kinds of activities from videos by an SVM multi-class classifier with a binary tree architecture. The framework is composed of three functionally cascaded modules: (a) detecting and locating people by non-parameter background subtraction approach, (b) extracting various of features such as local ones from the minimum bounding boxes of human blobs in each frames and a newly defined global one, contour coding of the motion energy image (CCMEI), and (c) recognizing activities of people by SVM multi-class classifier whose structure is determined by a clustering process. The thought of hierarchical classification is introduced and multiple SVMs are aggregated to accomplish the recognition of actions. Each SVM in the multi-class classifier is trained separately to achieve its best classification performance by choosing proper features before they are aggregated. Experimental results both on a home-brewed activity data set and the public Schüldt’s data set show the perfect identification performance and high robustness of the system.

Automatic video surveillance using statistical analysis of temporal posture sequences

PurposeThe problem of automatic recognition of human activity is one of the most important and challenging areas of research in computer vision because of the wide range of possible applications, for example surveillance, advanced human‐computer interactions, monitoring. This paper presents statistical computer vision approaches to automatically recognize different human activities.Design/methodology/approachThe human activity recognition process has three steps: firstly human blobs are segmented by motion analysis; then the human body posture is estimated and, finally a temporal model of the detected posture series is generated by discrete hidden Markov models to identify the activity.FindingsThe system was tested on image sequences acquired in a real archaeological site while some people simulated both legal and illegal actions. Four kinds of activity were automatically classified with a high percentage of correct detections.Research limitations/implicationsThe proposed approach provides efficient solutions to some of the most common problems in the human activity recognition research field: high detailed image requirement, sequence alignment and intensive user interaction in the training phase. The main constraint of this framework is that the posture estimation approach is not completely view independent.Practical implicationsResults of time performance tests were very encouraging for the use of the proposed method in real time surveillance applications.Originality/valueThe proposed framework can work using low cost cameras with large view focal lenses. It does not need any a priori knowledge of the scene and no intensive user interaction is required in the early training phase.

Temporal spatio-velocity transform and its application to tracking and interaction

This paper describes the temporal spatio-velocity (TSV) transform for extracting pixel velocities from binary image sequences. The TSV transform is derived from the Hough transform over windowed spatio-temporal images. We present the methodology of the transform and its implementation in an iterative computational form. The intensity at each pixel in the TSV image represents a measure of the likelihood of occurrence of a pixel with instantaneous velocity in the current position. Binarization of the TSV image extracts blobs based on the similarity of velocity and position. The TSV transform provides an efficient way to remove noise by focusing on stable velocities, and constructs noise-free blobs. We apply the transform to tracking human figures in a sidewalk environment and extend its use to an interaction recognition system. The system performs background subtraction to separate the foreground image from the background, extracts standing human objects and generates a one-dimensional binary image sequence. The TSV transform takes the one-dimensional image sequence and yields the TSV images. Thresholding of the TSV image generates the human blobs. We obtain the human trajectories by associating the segmented blobs over time using blob features. We analyze the motion-state transitions of human interactions, which we consider to be combinations of ten simple interaction units (SIUs). Our system recognizes the 10 SIUs by analyzing the shape of the human trajectory. We illustrate the TSV transform and its application to real images for human segmentation, tracking and interaction classification.