Portable Stereo Chute Box Imaging System Design and Development for Fisheries

It is important to obtain accurate fishing bycatch information to protect and manage marine wildlife and the environment regionally and globally. Fisheries have been dependent on self-reporting and/or observer programs for fishery data management. Self-reporting can be inaccurate, and deploying observers is time consuming and can only cover a small number of ships due to the high cost. Therefore, we need a cost effective, fast, and reliable data collection method that can be deployed across a wide coverage area. Electronic monitoring (EM) systems have been employed widely to collect image data. While EM systems are effective to collect data, data analysis including extracting and annotating fish data has historically been done manually. Post processing imagery by human observers is time consuming and costly. Recently artificial intelligence (AI) and machine learning (ML) have been applied to image data to identify fish species and estimate size and weight. These efforts can eventually lead to automatic catch, bycatch, and discard data collection using various sensors and vision systems. In this paper, we describe a portable stereo chute box system to collect depth information and color images of fish as a start to replacing human observers. Also, depth information from the system can be further used to innovate AI/ML algorithms including fish identification, and size and weight estimation. For the initial stage of the project, the stereo chute box system was placed over a conveyor belt, and color images and depth images were collected for analysis.