Hamdi A. Mahmoud

Fall detection system of elderly people based on integral image and histogram of oriented gradient feature

Falls represent a major cause of fatal injury, especially for the elderly, which accordingly create a serious obstruction for their independent living. Many efforts have been put towards providing a robust method to detect falls accurately and timely. This paper proposes an alerting system for detecting falls of the elderly people that monitors seniors via detecting the elderly faces and their bodies in order to generate an alert on falling detection. The proposed system consists of three phases that are pre-processing, feature extraction, and detecting phases. The integral image-based approach for multi-scale feature extraction developed to characterize the distinctive and robust patterns of different face poses. The histogram of oriented gradient (HOG) of extracted feature is then computed. The experiments were done on the datasets which consists of 191 recorded videos annotated human images with a large range of pose variations and backgrounds. The design of the fall detection system can increase the living time and reduce the rate of death due to the fall and shows the promising performance of the proposed system.

Cattle classifications system using Fuzzy K- Nearest Neighbor Classifier

This paper presents cattle classifications system using Fuzzy K- Nearest Neighbor Classifier (FKNN). The proposed system consists of two phases; segmentation and feature extraction phase and classifications phase. Expectation Maximization image segmentation (EM) algorithm was used to segments and extracts texture feature of each cattle muzzle image and their image color. Then, it followed by applying the FKNN for classification. The data sets used contains thirty two groups of cattle muzzle images. The experimental result proves the advancement of FKNN classifier better than other classification technique. FKNN achieves 100% classification accuracy compared to 88% classification accuracy achieved from K- Nearest Neighbor Classifier (KNN) classification system.

Action Recognition Using Fast HOG3D of Integral Videos and Smith-Waterman Partial Matching

Recognising human activity from video stream has become one of the most interesting applications in computer vision. In this study, a novel hybrid technique for human action recognition is proposed based on fast HOG3D of integral videos and Smith-Waterman partial shape matching of the fused frame. The proposed technique is divided into two main stages, the first stage extracts a set of foreground snippets from the input video, and extracts the histogram of 3D gradient orientations from the spatio-temporal volumetric data; and the second stage fuses a set of key frames from current snippet and extracts the contours from the fused frame. Non-linear support vector machine (SVM) decision trees are used to classify HOG3D features into one of fixed action categories. On the other hand, Smith-Waterman partial shape matching is used to compare between the contour of the fused frame and the stored template contour of specified action. The results from SVM and Smith-Waterman partial shape matching are then combined. The experimental results show that combining non-linear SVM decision trees of HOG3D features and Smith-Waterman partial shape matching of fused contours improved the accuracy of the classification model while maintaining efficiency in time elapsed for training.

Action recognition technique based on fast HOG3D of integral foreground snippets and random forest

Action recognition is considered a promising field in computer vision and can be used in many applications such as video indexing and retrieval. In this paper, we present a novel technique for action recognition based on traditional three stages of feature extraction, action learning, and action recognition. The proposed technique builds a foreground snippet from the input video file, then uses integral videos representation of the foreground snippet to extract HOG3D feature vector. After that, random forest is constructed and trained from feature space to classify the Weizmann actions. Several experiments are performed to show the effectiveness, invariance and speed of the proposed technique against state of the art techniques. Experiments are made on ten different human actions on the Weizmann dataset. The best obtained average recall and average specificity values were 95.68 and 93.21, respectively.

Weighted reduct selection metaheuristic based approach for rules reduction and visualization

Rules dimensionality may hinder the promising benefits of any automatic diagnostic approach. Recent studies pointed out toward meta-heuristics techniques due to their fast processing and promising results. In this paper, a hybrid approach combines the Rough Set Theory (RS) and the metaheuristic Genetic Algorithm (GA) to reach an efficient classifier of Breast Cancer disease with reduced rules. The proposed approach is called Rough set with Genetic Algoritm (RSGA). RS was used in RSGA for feature selection process, the reduct selection was performed using a weighted function based on the minimal mean square error of each reduct. The GA works for reducing extracted rules from the selected reduct. Microsoft Automatic Graph Layout (MSAGL) was used for visualizing reduced rules. The performance of the hybrid approach outperforms the results extracted from RS classifier and was evaluated using accuracy, specificity and sensitivity measures. RSGA achieved the highest performance for accuracy reached 98% and solved the problem of rules dimensionality and enabled visual information.

Sequential-based action recognition technique based on homography of interested SIFT keypoints

Automatic classification of human actions in video streams is considered one of the promising fields in computer vision, especially in the fields of video indexing, gesture recognition, and surveillance of sensitive assets. In this paper, we presented a sequential based approach for action recognition based on random sample consensus (RANSAC) matching of shape-based influential keypoints. The scale-invariant feature transform (SIFT) is used to provide an invariant descriptor for the action in the entire frame. Then a hypothesize matching is performed between the keypoints in reference and input frames using Normalized Cross Correlation (NCC). The initial matching from NCC is improved using RANSAC, which is used to find a consistent matching and to build the homography. Finally, Hausdorff distance is used for action recognition by measuring the closeness of the two sets. Experimental results show the effectiveness and accuracy of the proposed approach.

Intelligent road surface quality evaluation using rough mereology

The road surface condition information is very useful for the safety of road users and to inform road administrators for conducting appropriate maintenance. Roughness features of road surface; such as speed bumps and potholes, have bad effects on road users and their vehicles. Usually speed bumps are used to slow motor-vehicle traffic in specific areas in order to increase safety conditions. On the other hand driving over speed bumps at high speeds could cause accidents or be the reason for spinal injury. Therefore informing road users of the position of speed bumps through their journey on the road especially at night or when lighting is poor would be a valuable feature. This paper exploits a mobile sensor computing framework to monitor and assess road surface conditions. The framework measures the changes in the gravity orientation through a gyroscope and the shifts in the accelerometers indications, both as an assessment for the existence of speed bumps. The proposed classification approach used the theory of rough mereology to rank the modified data in order to make a useful recommendation to road users.