Multi-UAV to UAV Tracking

Slides:



Advertisements
Similar presentations
Learning Trajectory Patterns by Clustering: Comparative Evaluation Group D.
Advertisements

Road-Sign Detection and Recognition Based on Support Vector Machines Saturnino, Sergio et al. Yunjia Man ECG 782 Dr. Brendan.
Human Identity Recognition in Aerial Images Omar Oreifej Ramin Mehran Mubarak Shah CVPR 2010, June Computer Vision Lab of UCF.
Adviser : Ming-Yuan Shieh Student ID : M Student : Chung-Chieh Lien VIDEO OBJECT SEGMENTATION AND ITS SALIENT MOTION DETECTION USING ADAPTIVE BACKGROUND.
Large Lump Detection by SVM Sharmin Nilufar Nilanjan Ray.
Spatial Pyramid Pooling in Deep Convolutional
Jacinto C. Nascimento, Member, IEEE, and Jorge S. Marques
Jason Li Jeremy Fowers Ground Target Following for Unmanned Aerial Vehicles.
Object detection, tracking and event recognition: the ETISEO experience Andrea Cavallaro Multimedia and Vision Lab Queen Mary, University of London
3D SLAM for Omni-directional Camera
Object Detection with Discriminatively Trained Part Based Models
Supervised Learning of Edges and Object Boundaries Piotr Dollár Zhuowen Tu Serge Belongie.
Automated Target Recognition Using Mathematical Morphology Prof. Robert Haralick Ilknur Icke José Hanchi Computer Science Dept. The Graduate Center of.
Coin Recognition Using MATLAB - Emad Zaben - Bakir Hasanein - Mohammed Omar.
Yann LeCun Other Methods and Applications of Deep Learning Yann Le Cun The Courant Institute of Mathematical Sciences New York University
Cancer Metastases Classification in Histological Whole Slide Images
Video object segmentation and its salient motion detection using adaptive background generation Kim, T.K.; Im, J.H.; Paik, J.K.;  Electronics Letters 
Goal: Predicting the Where and What of actors and actions through Online Action Localization Figure 1.
Motion and optical flow
When deep learning meets object detection: Introduction to two technologies: SSD and YOLO Wenchi Ma.
CS 4501: Introduction to Computer Vision Object Localization, Detection, Semantic Segmentation Connelly Barnes Some slides from Fei-Fei Li / Andrej Karpathy.
Learning to Compare Image Patches via Convolutional Neural Networks
Convolutional Neural Fabrics by Shreyas Saxena, Jakob Verbeek
Krishna Kumar Singh, Yong Jae Lee University of California, Davis
A Neural Approach to Blind Motion Deblurring
Compositional Human Pose Regression
Introduction of Real-Time Image Processing
Introduction Computer vision is the analysis of digital images
CS 698 | Current Topics in Data Science
R-CNN region By Ilia Iofedov 11/11/2018 BGU, DNN course 2016.
Fast and Robust Object Tracking with Adaptive Detection
Fully Convolutional Networks for Semantic Segmentation
Scott Tan Boonping Lau Chun Hui Weng
A Convolutional Neural Network Cascade For Face Detection
Introduction to Neural Networks
Image Classification.
Learning to See in the Dark
Introduction Computer vision is the analysis of digital images
Vessel Extraction in X-Ray Angiograms Using Deep Learning
Introduction of MATRIX CAPSULES WITH EM ROUTING
Smart Robots, Drones, IoT
Object Detection Creation from Scratch Samsung R&D Institute Ukraine
Yang Liu, Perry Palmedo, Qing Ye, Bonnie Berger, Jian Peng 
Object Tracking: Comparison of
SIMPLE ONLINE AND REALTIME TRACKING WITH A DEEP ASSOCIATION METRIC
Introduction to Object Tracking
Introduction Computer vision is the analysis of digital images
CSC 578 Neural Networks and Deep Learning
VERY DEEP CONVOLUTIONAL NETWORKS FOR LARGE-SCALE IMAGE RECOGNITION
Object Detection Implementations
Multi-UAV Detection and Tracking
Moving Target Detection Using Infrared Sensors
Weak-supervision based Multi-Object Tracking
Report 7 Brandon Silva.
Report 4 Brandon Silva.
Appearance Transformer (AT)
Week 3 Volodymyr Bobyr.
SFNet: Learning Object-aware Semantic Correspondence
Volodymyr Bobyr Supervised by Aayushjungbahadur Rana
Report 2 Brandon Silva.
Self-Supervised Cross-View Action Synthesis
Multi-Target Detection and Tracking of UAVs from a UAV
Volodymyr Bobyr Supervised by Aayushjungbahadur Rana
Report 7 Brandon Silva.
Point Set Representation for Object Detection and Beyond
Robust Feature Matching and Fast GMS Solution
REU Program 2019 Week 6 Alex Ruiz Jyoti Kini.
Counting in High-Density Crowd Videos
Presented By: Firas Gerges (fg92)
Shengcong Chen, Changxing Ding, Minfeng Liu 2018
Presentation transcript:

Multi-UAV to UAV Tracking Brandon Silva, Waqas Sultani, Dr. Mubarak Shah, University of Central Florida Introduction Experiment Results continued UAVs (Unmanned Aerial Vehicles) are gaining in popularity for personal, commercial, and government use. Whether it be drone racing, autonomous delivery drones, or security drones; a fast and accurate collision avoidance system is needed as the airways become more cluttered. This project aims to create an end to end network for detection and tracking of multiple UAVs in videos. The best results obtained used these parameters: Input resolution of 640x640 Learning rate of 0.00006 with the Adam optimizer MSE loss, with a constant k=3 multiplied to entire loss function to weight detections in ground truth higher than pixel values of zero 20% test split: 40 videos for training and 10 for testing Ground truth is red, and predictions are green. Results Dataset The resulting chart shows testing F1-Scores as the model trains over 8 epochs. Different thresholds are used to determine how large of an output value should be to count as a detection, ranging from 0.2 to 0.8. Detections are counted as correct if their centers are within 5 pixels of the ground truth. This chart shows the number of false positives detected per epoch on the testing data. The best F1-Score was 78.46% with a false alarm rate of 0.2323. so far, and the model is improving with more fine tuning. This was calculated with a threshold of 0.2. Multi-Target Detection and Tracking from a Single Camera in Unmanned Aerial Vehicles (Purdue): 50 Videos, 30 FPS, 70250 frames Bounding box annotations around the UAVs Up to 8 UAVs in a single frame 1920x1080 and 1280x960 resolution Resized to 640x640 Method The UAV videos had extensive camera shake due to the nature of recording from a moving platform, so the input video is stabilized, using a Euclidean transformation and the Lucas-Kanade method for obtaining and matching key points between frames.  2. These stabilized frames are converted to grayscale and fed into a full 2D convolutional neural network that outputs a pixel level segmentation, where values close to 1 are considered UAV detections. This model is based on FoveaNet1. A middle frame is randomly selected from a video, then 2 consecutive frames before and after are stacked together and passed into the network. The network is trained on segmentations generated from the bounding boxes by calculating the center coordinate and filling in a circle around the point. These training labels can either be binary (1’s are detections) or gaussian (where a gaussian filter is applied about the center point). Empirically, the network performs similarly when trained on either label. 2D Conv Filters: 16 Kernel: 9 x 9 Max Pool Stride: 2 Kernel: 2 x 2 2D Conv Filters: 32 Kernel: 7 x 7 2D Conv Filters: 64 Kernel: 7 x 7 2D Conv Filters: 256 Kernel: 5 x 5 2D Conv Filters: 512 Kernel: 5 x 5 2D Conv Filters: 256 Dropout: 0.5 Kernel: 3 x 3 2D Conv Filters: 128 Dropout:0.5 Kernel: 3 x 3 2D Conv Filters: 1 Kernel: 1 x 1 Conclusion 1x320x320 Output Our method of solving the issue yieled a very reasonable F1-Score. The ridgeline and hills in the videos posed quite an issue with the network as it caused false detections in some of the videos, since all videos were recorded in the same area. Overall though, this end to end lightweight network for detecting these UAVs is shown to be a valid solution to this problem, without relying on optical flow or extensive preprocessing. This will further be enhanced with a novel tracking method to help reduce false alarms and increase the F1-Score. 5x640x640 Frames References 1. Lalonde, Rodney, et al. “ClusterNet: Detecting Small Objects in Large Scenes by Exploiting Spatio-Temporal Information.” 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, doi:10.1109/cvpr.2018.00421.