Vision-Based Unattended Terrace Children Safety Alert System

An unattended child can be subjected to a sudden fall from a terrace. The paper proposes a computer vision-based approach for detecting kids playing on the terrace. This system is implemented using a fusion of features extracted from HOG and SIFT (feature detectors). The extracted features were reduced using the k-means clustering technique. The Optimum number of clusters for the extracted features was calculated using the elbow method. The feature vector was further optimized using principal component analysis. Decision tree, KNN, Random Forest, and SVM were the algorithms used to determine the accuracy of the technique. The SVM-RBF gave a maximum accuracy of 85%.