Human detection in videos plays an important role in various real life applications. Most of traditional approaches depend on utilizing handcrafted features which are problem-dependent and optimal for specific tasks. Moreover, they are highly susceptible to dynamical events such as illumination changes, camera jitter, and variations in object sizes. On the other hand, the proposed feature learning approaches are cheaper and easier because highly abstract and discriminative features can be produced automatically without the need of expert knowledge. In this paper, we utilize automatic feature learning methods which combine optical flow and three different deep models (i.e., supervised convolutional neural network (S-CNN), pretrained CNN feature extractor, and hierarchical extreme learning machine) for human detection in videos captured using a nonstatic camera on an aerial platform with varying altitudes. The models are trained and tested on the publicly available and highly challenging UCF-ARG aerial dataset. The comparison between these models in terms of training, testing accuracy, and learning speed is analyzed. The performance evaluation considers five human actions (digging, waving, throwing, walking, and running). Experimental results demonstrated that the proposed methods are successful for human detection task. Pretrained CNN produces an average accuracy of 98.09%. S-CNN produces an average accuracy of 95.6% with soft-max and 91.7% with Support Vector Machines (SVM). H-ELM has an average accuracy of 95.9%. Using a normal Central Processing Unit (CPU), H-ELM's training time takes 445 seconds. Learning in S-CNN takes 770 seconds with a high performance Graphical Processing Unit (GPU).
This paper presents the concept of Simultaneous Localization and Multi-Mapping (SLAMM). It is a system that ensures continuous mapping and information preservation despite failures in tracking due to corrupted frames or sensor's malfunction; making it suitable for real-world applications. It works with single or multiple robots. In a single robot scenario the algorithm generates a new map at the time of tracking failure, and later it merges maps at the event of loop closure. Similarly, maps generated from multiple robots are merged without prior knowledge of their relative poses; which makes this algorithm flexible. The system works in real time at frame-rate speed. The proposed approach was tested on the KITTI and TUM RGB-D public datasets and it showed superior results compared to the state-of-the-arts in calibrated visual monocular keyframe-based SLAM. The mean tracking time is around 22 milliseconds. The initialization is twice as fast as it is in ORB-SLAM, and the retrieved map can reach up to 90 percent more in terms of information preservation depending on tracking loss and loop closure events. For the benefit of the community, the source code along with a framework to be run with Bebop drone are made available at https://github.com/hdaoud/ORBSLAMM.