Rochet et al : Precision and accuracy of fish length measurements obtained with two visual underwater methods 



" .'*^ 



ms' 



..Jg^l 3:02:26 



-*•- 



■. y 



^ 



Figure 3 



A METRAU video camera image with overlaid grid. 



camera has a built-in automatic focus unit which adjusts 

 lens settings to provide a sharp image. The camera is 

 remotely controlled by a RS323 digital link and sends 

 data back over this link, including zoom position and 

 focal value (see details in Cadiou et al., 2004). A previ- 

 ous calibration in air and in a test tank provided cor- 

 relation rules between raw data and field angle or focal 

 distance. 



When the system is operated, the data received by the 

 computer are processed in real time. The object distance 

 and the field angle are computed and a scale is overlaid 

 on the video image (Fig. 3). 



According to optical laws, the depth of focus decreases 

 as the focal length increases. This means that in order 

 to obtain an accurate measure of focal distance, a nar- 

 row field angle is required. In addition, the depth of 

 focus increases when the focal distance moves towards 

 infinity. Consequently, a domain of validity of the mea- 

 surement can be defined. With the METRAU camera, 

 there must be a target distance under three meters and 

 a field angle of less than 6°. These constraints have to 

 be combined with the following conditions: a steady 

 image that would allow the automatic focal servo to 

 stabilize; and avoidance of scenes with several image 

 planes. In turbid waters, particles can create disturbing 

 focal planes and affect the measurement process. 



Measurement experiments 



Artificial objects and live fish Objects of known size 

 were used to estimate the potential bias in the length 

 measurements obtained with the two devices. Three 

 rigid objects — a can, a bottle, and a plastic tube mea- 

 suring respectively 13, 30, and 66 cm — were repeatedly 

 measured to evaluate device performance and observer- 

 induced variability in the absence of errors induced by 

 fish movement and variations in horizontal observation 

 angles. 



Fish movement makes the horizontal observation 

 angle vary continuously. As a result, it is difficult to 

 judge if and when an individual fish is perpendicular 

 to the measurement axis. Further, fish seldom lie in 

 a straight plane. Some species continuously flex their 

 tail, others bend their whole body. To mimic the mobil- 

 ity of a real fish, a mobile object was built consisting 

 of several pieces of Ertalyte (Quadrant Engineering 

 Plastic Products, Bridgeport. CT) plates linked together 

 with rope rings. This artificial fish was designed to be 

 neutrally buoyant so that it could be moved by water 

 currents and undulate like a real swimming fish. The 

 "artificial fish" had three distinctively colored parts. 

 Thus depending on how many parts were measured, a 

 small (13 cm), medium (17 cm), or large (41 cm) "artifi- 

 cial fish" was the result. The real size of rigid objects 

 and of the artificial fish was unknown to the observers 

 throughout the measurement experiment. 



In addition to measuring each of the rigid objects 

 and the artificial fish, 351 individuals belonging to 21 

 deep-sea fish species were measured with both methods. 

 The body sizes of these species ranged from 5 to 110 cm. 

 Each individual fish was measured up to nine times. 

 Altogether 2373 measurements were carried out. 



Real time and postoperation measurements While the 

 ROV was in operation, four to five observers were able 

 to watch the video images. Real time measurements 

 were performed by estimating sizes directly from the 

 screen, without using any measuring instrument. Each 

 observer was asked to write down his or her length 

 estimate without announcing it, so that independent 

 measurements were obtained. All artificial objects were 

 measured by both trained and novice observers; real fish 

 were measured only by trained observers, namely scien- 

 tists and ROV pilots. All objects and fish were measured 

 at distances of 2 to 5 meters. 



Postoperation measurements were also performed 

 on registered videos and digital images. For the laser 

 method, the video tape was replayed. The tape was 

 stopped when the image with an object or fish seemed 

 to be in the best possible position. The fish or object was 

 then measured with a ruler on the still video image. 

 Postoperation measurements made with a ruler were 

 also performed on digital snapshots taken from the 

 videos in real time for both the laser and the METRAU 

 method. A ruler was used rather than computer image 

 analysis because it was easy and cost-efficient and it 

 was felt appropriate for this trial appraisal of measure- 

 ment methods. The bias introduced by this method was 

 assumed to be negligible compared to observer-induced 

 and fish-movement-induced errors. 



Operational constraints prevented a full factorial 

 design where all observers could use all methods and 

 measure all objects. 



Data analysis 



Variance components for observers and fish movements 



The measurement variability due to observer differences 



