Application Requirement-Driven Automatic ISP Parameter Tuning for a Rear View Monitoring Camera

Abstract

An image signal processor (ISP) is a dedicated processor that transforms raw data obtained from camera sensors into an image that satisfies the requirements of a specific application or use case. An ISP typically has many tuning parameters due to the complexity of the image transformation. Until now, these are generally tuned by human experts manually, and this work takes a great deal of time. This paper proposes an application-level automatic ISP parameter tuning system. In particular, this paper focuses on a rear view monitoring (RVM) camera that is mounted on the rear side of the vehicle to prevent backovers in advanced driving assistance systems (ADAS). The proposed system consists of four steps. The first step is the input image generation, which captures a virtual scene including the test site and vehicle body whose three-dimensional (3D) models are created according to the test requirements and vehicle design. The second step is ISP processing, which transforms the input image into an ISP output image (RVM image) according to the ISP specification in the RVM system. The third step is to evaluate the RVM image’s fitness using evaluation criteria (EC) functions. Finally, ISP parameters are tuned by the grid-to-random search method. In the experiment, the proposed system is evaluated by using the 3D modeling data of six different test vehicle types. Experimental results show that the proposed system can effectively obtain the usable ISP parameters’ values that satisfy all RVM requirements for a given situation, regardless of test vehicle type, within 2~3 hours.

Keywords

• Automatic parameter tuning
• grid-to-random search
• image signal processor (ISP)
• rear view monitoring camera