A Novel Framework for Detection of Objects from Video Using Deep Learning

Detection of objects in the videos is a most challenging task in surveillance, autonomous driving, robotics, etc. Traditional approaches for detection of objects in videos rely on extraction of features and image processing, which requires a significant amount of manual intervention and having complex and cluttered environments. These algorithms in surveillance systems lack accuracy and most work with labeled samples. The algorithms are time consuming and take significant time for object detection. Nowadays, machine learning is being used mostly for detection of moving objects in videos. Objects detection techniques based on deep learning are promising in the surveillance systems, but they are facing some limitations and drawbacks that attract the attention of the researchers. These deep learning-based algorithms require large amounts of computation, delay detection and tracking, and may not work in novel environments or object categories where training data is not available. This paper demonstrates a novel deep learning method for detection of various objects in videos, for e.g., CCTV surveillance systems. The proposed framework is highly effective in surveillance systems because it can detect and track objects accurately and quickly in real time. The recognition of the moving object has been detected using LBPH (Local Binary Pattern Histogram), eigenface, and Fisherface algorithms. All these algorithms are implemented and compared using different parameters such as accuracy, precision, recall, and F1 score. The results show that the LBPH algorithm enhances accuracy than the other two algorithms. The proposed system achieves up to 89% of accuracy for detection of objects using LBPH, which is 8–12% more than the eigenface and Fisherface algorithms.