Search

In our research we compare various neural network architectures that are used for object detection and recognition. In this work vehicles and pedestrians are considered objects of interest. Modern artificial neural networks are able to detect and localize objects of known classes. This allows them to be used in various technical vision systems and video analysis systems. In this paper we compare three architectures (YOLO, Faster R-CNN, SSD) by the following criteria: processing speed, mAP, precision and recall.

The object tracking task for simultaneous video capturing in visual and infrared spectral ranges is considered. The tracking algorithm based on independent processing of images that were captured in different ranges and tracking result fusion is proposed. The preferred spectral range selection based on performance indicator analysis is suggested. The experimental examination results are given.

In this paper we are considering road situation analysis tasks for traffic control and ensuring safety. The following image processing algorithms are proposed: vehicle detection and counting algorithm, road marking detection algorithm. The algorithms are designed to process images obtained from a stationary camera. The developed vehicle detection and counting algorithm was implemented and tested also on an embedded platform of smart cameras. The results of experimental research of proposed algorithms are presented.

Lines are one of the most informative structure elements on any images. For this reason, objects detection and recognition problems are often reduced to edge detection task. One of the most popular approaches to detect lines is based on the Hough transform or Radon transform. However, using both of transforms allows estimating the infinite lines parameters only. It is necessary to use additional approaches to estimate the ends of the detected lines. Moreover, Radon transform does not allow detecting non-straight curve shapes at all. This work is oriented to solve line detection problem using Radon transform and multi-agent approach. The results of the experimental researches that confirm the effectiveness of the proposed approach are given. The real full HD image sequences are used. The direction of further improvements is proposed.

This paper describes the implementation of the orientation estimation algorithm in FPGA-based vision system. An approach to estimate an orientation of objects lacking axial symmetry is proposed. Suggested algorithm is intended to estimate orientation of a specific known 3D object based on object 3D model. The proposed orientation estimation algorithm consists of two stages: learning and estimation. Learning stage is devoted to the exploring of studied object. Using 3D model we can gather set of training images by capturing 3D model from viewpoints evenly distributed on a sphere. Sphere points distribution is made by the geosphere principle. Gathered training image set is used for calculating descriptors, which will be used in the estimation stage of the algorithm. The estimation stage is focusing on matching process between an observed image descriptor and the training image descriptors. The experimental research was performed using a set of images of Airbus A380. The proposed orientation estimation algorithm showed good accuracy in all case studies. The real-time performance of the algorithm in FPGA-based vision system was demonstrated.

This paper describes the implementation of the multiple targets tracking algorithm in FPGA-based vision system. The described algorithm was designed to process such situations as the object trajectories crossing and the temporary object screening by other objects. The source data for this algorithm is a list of the parameters of the previously extracted binary regions from each frame of the sequence. The main idea of this algorithm is to represent the source data as a bipartite graph and split it into insolated elementary graphs corresponding to five situations: object is moving or staying still, a new object detected, object is missed, the pair of the objects is merged into one and the region is divided. These graphs are used to form a new object list. The goal of this work was to implement the described algorithm in small-sized onboard vision system based on the single Xilinx FPGA using MicroBlaze soft processor block. In the proposed implementation of this algorithm recursive procedures were replaced with table-based procedures. The experimental research of the algorithm shows the increasing tracking performance 5 – 9 times on previously described hardware.

We analyze the registration process in combined vision systems and offer a new approach for image synthesis. The proposed algorithm contains all benefits of previously developed approaches. Compared to the other approaches, it involves two techniques: searching and tracking. Tracking technique is performed by using feature detection and matching algorithms. Searching technique consists of correlation-extremal algorithm. We describe the advantages of the presented idea in experimental results, give some examples which reflect the speed-up and the quality of registration and explain their joint applying.

In this paper we propose an object tracking approach that includes the selection of the appropriate coordinate estimation method from the set of base methods: cross-correlation, statistical segmentation, methods based on spatial and spatiotemporal filtering. The tracking performance of the base methods is estimated by means of the several parameters (performance features) that describe the reliability of tracking. The features are invariant to the changes of the mean brightness, contrast, scale and rotation. The comparison of the performance of the methods is based on the binary performance characteristic that describes the tracking in the terms of work/not work. The proposed object tracking algorithm is implemented in five steps which are carried out on each frame: the calculation of the performance feature for each base coordinate estimation method, the calculation of the binary performance characteristic for each method, the selection of the appropriate method, the estimation of the object coordinates using the selected method, the correction of the tracking process performed by other methods. The algorithm shows a good tracking performance during the variations of the observation conditions. In most cases it comprises the main advantages of each of the source methods.

In this work an algorithm of the tracking of the set of moving objects is described. The important features of the task are crossings of the object trajectories and temporary screening of the objects by other objects. The source data for the proposed algorithm is a list of the parameters of the binary regions extracted from each image of the sequence. The main idea of the considered algorithm is to build a bipartite graph. The recoursive procedure is used to partition the graph into connected graphs corresponding to five situations: detection of a new object, missing object, merging of the objects into one region, division of the region and "simple" object tracking. These graphs are used to form a new list of the objects. The experimental research of the algorithm shows a good tracking performance in both ground and aerial environments.

In the practice of line detection applications based on Radon transform (RT), noise and clutter decrease the sharpness of the RT local peaks, which correspond to linear edges in the original image. We suggest a new approach to line detection on gray-scale images using the so-called weighted Radon transform (WRT). The proposed WRT-based approach exploits gradient direction information in such a way that only the gradient component perpendicular to the line direction is integrated to form a local peak corresponding to the line. Theoretical and experimental studies have shown the effectiveness of the suggested WRT-based line detection method.

One of the most popular approaches to detect lines is based on the Hough transform. Radon transform (RT) allows to made Hough transform by FFT. But real-world difficulties (such as: noise, clutter, blur etc.) decrease the sharpness of the local maximums of RT. The new approach to line detection using the Weighted Radon Transform (WRT) is suggested in this paper. This approach uses gradient direction information, so only the differences that are perpendicular to the line direction are integrated to make a local maximum corresponding to the line. The theoretical and experimental studies show the effectiveness of the WRT-based line detection.