Refine your search
Collections
Co-Authors
Year
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z All
Aleksandrovich Kazantsev, Pavel
- Head-Shoulder Detection using Deep Autoencoders
Abstract Views :139 |
PDF Views:0
Authors
Pavel Vyacheslavovich Skribtsov
1,
Pavel Aleksandrovich Kazantsev
1,
Aleksey Vladimirovich Dolgopolov
1
Affiliations
1 PAWLIN Technologies Ltd, Dubna, Russia
1 PAWLIN Technologies Ltd, Dubna, Russia
Source
Indian Journal of Science and Technology, Vol 9, No 42 (2016), Pagination:Abstract
Background/Objectives: The main objective of this work is to create a new head-shoulder detection algorithm that is able to perform robustly with arbitrary camera's point of view and in any environment in real time on the CPU. Method: To achieve the objective the developed head-shoulder detection algorithm uses two cascades: HOG-Cascade and Cascade based on automatically extracted features. Deep neural network autoencoder is applied for automatic feature extraction. In this work, the positive examples were extracted from the TUD Multiple View Pedestrian database. The negative images that do not contain people were taken from the Nicta Dataset. We also used positive and negative examples collected from online surveillance cameras. The dataset was divided into training, test and validation. Detection quality was estimated using the Recall and False Positives Per Image (FPPI) metrics. Findings: Our study has shown that the combined use of HOG-features and automatically extracted features allows achieving high quality (Recall: 85%, FPPI: 0.5) and performance (5.5 FPS at CPU) in the problem of head-shoulder detection. The results demonstrate that the proposed method is comparable or slightly outperforms the state-of-the-art methods. Improvements/Applications: In the application of the developed detector to real video, it would be possible to reduce the level of false positives and increase the performance using the motion mask or filter based on the model of the scene.Keywords
Adaboost, Deep Autoencoders, Head-Shoulder Detection, HOG, Naive-Bayes Classifier- Ship Detection in Images Obtained from the Unmanned Aerial Vehicle (UAV)
Abstract Views :148 |
PDF Views:0
Authors
Aleksey Vladimirovich Dolgopolov
1,
Pavel Aleksandrovich Kazantsev
1,
Nikolay Nikolaevich Bezuhliy
1
Affiliations
1 Nakhimov Black Sea Higher Naval School, Sevastopol, Russia
1 Nakhimov Black Sea Higher Naval School, Sevastopol, Russia
Source
Indian Journal of Science and Technology, Vol 9, No 46 (2016), Pagination:Abstract
Background/Objectives: The main objective of this work is the creation of a new algorithm for detecting ships in the high resolution images obtained by unmanned aerial vehicle. Method: To achieve the objective the developed ship detection algorithm uses two cascades: shape-based cascade and cascade based on automatically extracted features. Deep neural network autoencoder is applied for automatic feature extraction. Findings: Images from Google Maps, Yandex Maps, and a small fraction of the images taken from the Internet were used for training and testing the developed approach. In total, the dataset consists of 1000 images containing ships. On average, each image contains 5.1 ships. The dataset was divided into training (50%) and test (50%). Detection quality was estimated using the Recall and Precision metrics. Our study has shown that the combined use of shape-based cascade and cascade based on automatically extracted features allows achieving high quality (Recall: 0.95, Precision: 0.94) and performance in the problem of ship detection by UAV. The results demonstrate that the proposed method is comparable or slightly outperforms the state-of-the-art methods. Improvements/Applications: Applying the detector in real video from the UAV, it would be possible to improve performance by parallelization of this approach on the GPU.Keywords
Connected Components, Deep Autoencoders, Naive-Bayes Classifier, Ship Detection, UAV.- Fast Multi-Object Tracking-by-Detection Using Tracker Affinity Matrix
Abstract Views :185 |
PDF Views:0
Authors
Affiliations
1 PAWLIN Technologies Ltd., Dubna, RU
1 PAWLIN Technologies Ltd., Dubna, RU
Source
Indian Journal of Science and Technology, Vol 9, No 27 (2016), Pagination:Abstract
Objectives: This study is an attempt to develop a fast real-time multi-object tracking algorithm with competitive precision and accuracy in crowded scenes. Methods/Statistical Analysis: This research extends multi-object tracking-by-detection framework by utilizing physical characteristics of trackers and their affinity for tracker guidance, in addition to detections and representation models. Guidance is done by particle filter which weights are updated through reworked observation model, featuring a term based upon tracker affinity. Tracker affinity is also used for tracker grouping that removes redundant trackers following the same target, and triggering a special propagation mode during occlusions that prevent identity switches. Findings: Our research has shown that trackers affinity matrix and algorithm features based on it yield substantial extra accuracy for tracking-by-detection framework, while taking a minor fraction of processing time. Improvements/Applications: In this paper we have introduced an approach that makes it possible to use less accurate, but faster detectors and representation model classifiers, enabling real-time processing, while keeping competitive precision and accuracy.Keywords
Multi-Object Tracking, Tracking-by-Detection, Particle Filter, Pedestrian Detection, Particle Filtering, Online Learning, Tracker Affinity, Surveillance.- Multi-Layer Neural Network Auto Encoders Learning Method, using Regularization for Invariant Image Recognition
Abstract Views :143 |
PDF Views:0
Authors
Affiliations
1 PAWLIN Technologies Ltd., Dubna, Moscow Region, RU
1 PAWLIN Technologies Ltd., Dubna, Moscow Region, RU