Day 4

Detailed paper information

Back to list

Paper title Using VENµS SuperSpectral Camera (VSSC) for moving vehicle detection
  1. Manuel Salvoldi Ben-Gurion University of the Negev Speaker
  2. Aviv L. Cohen-Zada Ben-Gurion University of the Negev
  3. Arnon Karnieli The Jacob Blaustein Institutes for Desert Research, Ben-Gurion University
Form of presentation Poster
  • B6. National missions TPM
    • B6.01 National EO satellite missions
Abstract text Monitoring transportation for planning, management, and security purposes in urban areas has been growing in interest and application by various stakeholders. Since the late 1990s, commercial very-high-resolution (VHR) satellites have been used for developing vehicle detection methods, a domain previously governed by aerial photography due to superior spatial resolution. Despite the apparent advantages of using air- or drone-borne systems for vehicle detection, several methods were introduced in the last two decades, utilizing space-borne VHR imagery (e.g., QuickBird, WorldView-2/3) with meter (multispectral bands) to submeter (panchromatic band) resolutions. Several of the applications applied machine learning for identifying parking cars. However, for detecting moving cars, two sensor capabilities have been utilized: (1) stereo mode by either satellite constellation or by body pointing abilities; and (2) a gap in the acquisition time between the push-broom detector sub-arrays. Changes in the location of moving objects can be observed between image pairs or across spectral bands, respectively. Both cases require overcoming differences in ground sampling distance and/or prerequisite spectral analyses to identify suitable bands for change detection.
Since January 2018, new multispectral products have been available for the scientific community provided by the Vegetation and Environmental New Micro Spacecraft (VENµS). This mission is a joint venture of the Israeli Space Agency (ISA) and the French Centre National d’Etudes Spatiales (CNES). The overall aim of the VENµS scientific mission is to acquire frequent, high-resolution, multispectral images of pre-selected sites of interest worldwide. Therefore, the system is characterized by the spatial resolution of 5 m per pixel (the upcoming mission phase will increase the spatial resolution to 4 m per pixel), the spectral resolution of 12 narrow bands in the visible-near infrared regions of the spectrum, and revisit time of 2 days in the same viewing and azimuth angles.
Here we demonstrate the VENµS capability to detect moving vehicles in a single pass with a relatively low spatial resolution. The VENµS Super Spectral Camera (VSSC) has a unique stereoscopic capability since two spectral bands (number 5 and 6), with the same central wavelength and width (620 nm and 40 nm, respectively), are positioned at extreme ends of the focal plane (Figure 1). This design enables a 2.7-sec difference in observation time. We took a straightforward approach to create a simple spectral index for moving vehicles detection (MVI) using these bands. Since the two bands are identical, there is no need for prior image analyses for dimensionality reduction or geometric corrections, as required for other sensors. Each moving vehicle is represented by a pair of bright blob-shaped clouds of pixels on a darker background (Figure 2). The center of each cloud in the pair is determined based on the same methodology used to identify the barycenter of a multi-particle system, where the MVI values replace the masses of the particles. Once the center of each cloud is known, the velocity vector, i.e., speed magnitude and orientation, can be extracted by geometrical considerations.
Results show successful detection of moving small- to medium-size vehicles. Especially interesting is the detection of private cars that are on average 2-3 m smaller than the ground sampling distance of VENµS. We effectively detected vehicle movement in different backgrounds/environments, i.e., on asphalt and unpaved roads, as well as over bare soil and plowed fields, and at different speeds, e.g., 61 km/h for a car over an asphalt road, and 19 km/h for vehicles on unpaved road. A speed of 111 km/h was calculated for a heavy train. This speed is in line with the engine speed limit and the regulations applied by Israeli authorities, providing estimation for MVI accuracy.
The MVI benefits from the coupling of a unique detector arrangement of the Super Spectral Camera onboard VENµS. In addition, the very high temporal resolution of 2 days makes VENµS products an attractive input for vehicle detection applications, particularly for operations that require monitoring on a nearly daily basis. It appears to be cost-effective compared to VHR commercial satellite and complex UAV-base monitoring systems. Furthermore, the MVI suggests that such bands arrangement is highly effective and should be considered for future space missions, primarily for surveillance and transportation monitoring.