Image Quality Standards in Automotive Vision Applications

Abstract

Digital cameras are increasingly used in automotive applications. As these cameras are integrated into active safety systems, image quality becomes ever more important. The captured image information is limited not only by the sensitivity and signal-to-noise ratio of the image sensor, but also by such image capture conditions as ambient lighting, camera-object distance, and relative camera-object velocity. For example, the detection of a pedestrian at night might be further impeded by the headlights of oncoming traffic. Image quality performance characterizes the ability of the imaging system to capture vital image information under application-typical capture conditions. Characterizing objective image quality performance requires clearly defined image quality attributes and metrics. This paper introduces the basic concepts of objective and subjective image quality, reviews existing image quality metrics and standards that have been developed for digital still and video applications, and explores their applicability for automotive uses where image information is interpreted by a human observer or machine vision application. The automotive photospace is introduced as a useful tool to characterize the automotive image capture conditions, and distinguish them from other still and video applications. As automotive imaging becomes more widespread, early standardization of image quality is important. This will enable automotive camera suppliers and the automotive industry to communicate in a common language when specifying imaging systems so that sufficient image quality under application conditions is ensured.

No-Reference Image Quality Assessment Based on the Fusion of Statistical and Perceptual Features

“…algorithms heavily depends on the image quality [ 6 , 7 ]. As a consequence, monitoring image/video quality is also crucial in vision-based advanced driver-assistance systems [ 8 ].…”

Fusion of thermal and visible images for day/night moving objects detection

“…In [18], authors used it on histograms of the RGB colour space. In other works, a luminance based method [19] and contrast metrics [20] were tested. Here, we present two approaches based on the HSV and CIELab colour spaces respectively.…”

An adaptive multiple-reset CMOS wide dynamic range imager for automotive vision applications

“…The probability density function of the luminance levels at which automotive scene features occur is of special interest, and initial photospace distributions for automotive day and nighttime traffic conditions have been reported earlier [2]. 3 shows a typical night traffic scene.…”

“…iSNR can be determined from laboratory measurements: the response curve D(L) is the opto-electronic conversion function (OECF), and the incremental gain its local gradient. ISO 15739 [4] proposes to measure noise simultaneously to OECF on a test chart, and OECF-noise measurements on a WDR transmission step chart have been demonstrated in previous publications [2], [5].…”