Muhammad Saqib

A study on detecting drones using deep convolutional neural networks

The object detection is a challenging problem in computer vision with various potential real-world applications. The objective of this study is to evaluate the deep learning based object detection techniques for detecting drones. In this paper, we have conducted experiments with different Convolutional Neural Network (CNN) based network architectures namely Zeiler and Fergus (ZF), Visual Geometry Group (VGG16) etc. Due to sparse data available for training, networks are trained with pre-trained models using transfer learning. The snapshot of trained models is saved at regular interval during training. The best models having high mean Average Precision (mAP) for each network architecture are used for evaluation on the test dataset. The experimental results show that VGG16 with Faster R-CNN perform better than other architectures on the training dataset. Visual analysis of the test dataset is also presented.

Drone-vs-Bird detection challenge at IEEE AVSS2017

Small drones are a rising threat due to their possible misuse for illegal activities, in particular smuggling and terrorism. The project SafeShore, funded by the European Commission under the Horizon 2020 program, has launched the “drone-vs-bird detection challenge” to address one of the many technical issues arising in this context. The goal is to detect a drone appearing at some point in a video where birds may be also present: the algorithm should raise an alarm and provide a position estimate only when a drone is present, while not issuing alarms on birds. This paper reports on the challenge proposal, evaluation, and results1.

Detecting dominant motion patterns in crowds of pedestrians

As the population of the world increases, urbanization generates crowding situations which poses challenges to public safety and security. Manual analysis of crowded situations is a tedious job and usually prone to errors. In this paper, we propose a novel technique of crowd analysis, the aim of which is to detect different dominant motion patterns in real-time videos. A motion field is generated by computing the dense optical flow. The motion field is then divided into blocks. For each block, we adopt an Intra-clustering algorithm for detecting different flows within the block. Later on, we employ Inter-clustering for clustering the flow vectors among different blocks. We evaluate the performance of our approach on different real-time videos. The experimental results show that our proposed method is capable of detecting distinct motion patterns in crowded videos. Moreover, our algorithm outperforms state-of-the-art methods.

Texture-based feature mining for crowd density estimation: A study

Texture feature is an important feature descriptor for many image analysis applications. The objectives of this research are to determine distinctive texture features for crowd density estimation and counting. In this paper, we have comprehensively reviewed different texture features and their different possible combinations to evaluate their performance on pedestrian crowds. A two-stage classification and regression based framework have been proposed for performance evaluation of all the texture features for crowd density estimation and counting. According to the framework, input images are divided into blocks and blocks into cells of different sizes, having varying crowd density levels. Due to perspective distortion, people appearing close to the camera contribute more to the feature vector than people far away. Therefore, features extracted are normalized using a perspective normalization map of the scene. At the first stage, image blocks are classified using multi-class SVM into different density level. At the second stage Gaussian Process Regression is used to re gress low-level features to count. Various texture features and their possible combinations are evaluated on publicly available dataset.