[Papers] Automatic Self-Improvement Scheme in Optical Flow-Based Motion Estimation for Sequential Fisheye Images

Abstract

the object tracking, estimating 3D structure, video compression, video frame-rate conversion, and video surveillance system. The motion estimation technique forecast the apparent motion in time-varying images that obtained from the projection process of the realworld one 1). This motion occurs because of exact object motion, camera motion, an illumination change, and noise. The objective of the motion estimation technique is then to define a motion vector for each pixel or group of pixels in an image by using a specific calculation scheme. This calculation scheme involves the observed image and at least one of its neighbors. To the extent our observation, at least there are two types of motion estimation scheme that have been developed massively in recent years 1). The first is optical flow-based motion estimation that calculates the apparent motion vectors directly from two successive images. The calculation performs to solve the brightness constancy assumption between pixels in an image and each counterpart pixel in the following or previous image. The optical flow is very well known in computer vision applications. Some of them are mentioned above. The second one is block-based motion estimation. This concept directly calculates the similarity of each block of pixels in an image with each counterpart block of pixels in one of the following or previous images. Therefore, a motion vector can be obtained for each block. The blockbased motion estimation is very valuable in the video compression application. Besides the two major motion estimation concepts, other schemes have been developed in recent years 1). Among these, there is the global motion estimation which deals with a video sequence that contains camera motion. This motion then is modeled by parametric transformations. How to estimate the parametric transformations is known as the global motion estimation. This motion estimation is essential for the sports video content analysis, including zoom in and out applications. Another method is the motion estimation in the transform domain. This method utilizes correspondence between translations in the spatial domain with a change of phase of the Fourier coefficients. Therefore, the motion is estimated from a Abstract This paper aims to present an innovative design of motion estimation for sequential fisheye images. This design is an extended version of the original Lucas and Kanade's (LK) concept that used to design for calculating optical flow from general perspective images. The extended design consists of the LK concept and an additional self-improvement mechanism that automatically finds the maximum performance of the estimated motion. This extended scheme works much better than the original LK's idea or some block-based motion estimations. Moreover, to some extent, this proposed method is working extremely well to overcome some critical characteristics of the sequential fisheye images. These characteristics include distortion error on the fisheye image area, inconsistent brightness level, fluctuating numb...

Sistem Pendeteksi Pejalan Kaki Di Lingkungan Terbatas Berbasis SSD MobileNet V2 Dengan Menggunakan Gambar 360° Ternormalisasi

“…Menggunakan Gambar 360° Ternormalisasi (Nafisun Nufus) 129 F. Insta 360° One R Pada penelitian ini kamera yang digunakan adalah kamera Insta 360° one R yang sering disebut dengan kamera 360° yang berfungsi untuk mendeteksi objek berupa pejalan kaki dengan keadaan secara real-time dan offline. Pada dasarnya teknologi kamera 360° adalah kamera yang memungkinkan untuk mengambil gambar dari semua sudut (360°) sehingga mampu menghasilkan video yang dapat dilihat dari berbagai sisi-depan, belakang, samping, atas dan bawah [10]. Pada Gambar 5 hasil dari pengambilan data menggunakan kamera 360° yang umumnya kamera 360° mempunyai dua lensa yaitu lensa depan 180° (gambar bagian atas) dan lensa belakang 180° (gambar bagian bawah) yang ternormalisasi untuk mengambil dan menggabungkan gambar secara bersamaan.…”

Sistem Pendeteksi Pejalan Kaki di Lingkungan Terbatas Berbasis SSD Mobilenetv1 Menggunakan Gambar 360° Ternormalisasi

“…Dengan NVIDIA Jetson AGX Xavier sekarang pengembang dapat menjalankan framework dan model AI untuk aplikasi pengenalan gambar, pendeteksian objek, segmentasi, pengenalan suara dan masih banyak lagi [6] F. Kamera 360° Pada penelitian ini menggunakan kamera 360° yang berfungsi untuk mendeteksi objek berupa pejalan kaki dengan keadaan secara real-time. Pada dasarnya teknologi kamera 360° adalah kamera yang memungkinkan kita untuk mengambil gambar dari semua sudut (360°) sehingga mampu menghasilkan video yang dapat dilihat dari berbagai sisi depan, belakang, samping, atas dan bawah [7]. Umumnya kamera 360° mempunyai dua lensa yang mampu tersinkronisasi untuk mengambil dan menggabungkan gambar secara bersamaan.…”

An Elastic Frame Rate Up-Conversion for Sequential Omnidirectional Images

“…The performance of these methods has yet to be tested in the case of videos taken from omnidirectional cameras. In comparison to the conventional video, image frames inside this omnidirectional video suffer a lot of distortion [15]- [21]. The most significant distortion comes from the radial distortion, and it is discernible clearly by human vision.…”

“…As depicted in Fig. 1 [21], the shape of a moving object, a cube in the first row (from a to c) and a train in the second row (from d to f), looks to be different while the omnidirectional camera captured and projected them on three different locations of the omnidirectional image (at the left side, at the center, and the right side). Even though the cube images are synthetic omnidirectional images, while the train images are the real ones, they perform identical behavior.…”

“…Moreover, in the central area, the shape of the object deformed massively. This is due to the construction of the omnidirectional lens that causes such inconsistent deformation [15], [17], and [21]. For example, Fig.…”