Advances in horse morphometric measurements using LiDAR

1 Advances in horse morphometric measurements using 1 LiDAR 2 3 M. Pérez-Ruiz1,2; D. Tarrrat-Martín1; M. J. Sánchez-Guerrero3; M. Valera3 4 5 1. Universidad de Sevilla. Área de Ingeniería Agroforestal 6 Dpto. de Ingeniería Aeroespacial y Mecánica de Fluidos, Spain. 7 2. Corresponding Author’s contact information: 8 ETSIA. Ctra. Sevilla-Utrera km 1 9 Sevilla, Spain 41013 10 Phone: 954 481 389 11 Fax: 954 486 436 12 E-mail: [email protected] 13 3. Universidad de Sevilla. Dpto. de Ciencias Agroforestales, Spain 14 15 16 Abstract. 17 Zoometric measurements have a potential value in differentialing between individuals and within 18 populations. The measurement of body size in horses and livestock plays a significant role in 19 functional longevity, production, and reproductive performance and health. In this context, the 20 measurements obtained without contact by detection systems and visualized by computers 21 could represent a great advance over conventional measurements that are tedious, time 22 consuming and stressful for the animals. This study presents a new approach to taking 23 zoometric measurements of an animal's body based on digital three-dimensional modelling. The 24 capture of the data series was carried out by a LiDAR sensor. The 16 laser beams of the sensor 25 were able to fully scan a horse, performing a 3D reconstruction of the horse's side, through 26 which body measurements were obtained. Five Pura Raza Española horses (PRE) (3 stallions 27 and 2 mares) with ages ranging between 5 and 18 years old were scanned. The PRE is the 28 most recognized native Spanish horse population for its census (national and international), 29 2 cultural and socioeconomic importance. For each horse, 17 zoometric measurements (linear 30 and angular) were taken both manually and using the LiDAR-based system to check the 31 usefullness of this non-invasive technology in obtaining quick livestock measurements while 32 causing minimal stress to the animals. Of the 17 zoometric measurements obtained manually 33 and with the sensor, 10 (58.82%) had a mean relative error that ranged between > 0 and <10; 5 34 (29.41%) had an error that ranged ≥10 and <20; and two (As and ACr) had an error ≥20 35 (11.76%). A total of 82.5% of the traits studied had an accuracy (v2) lower than 5%. Therefore, 36 although this approach could still be improved, it verifies the viability of noncontact 37 measurements of large livestock. 38 39 Keywords. Zoometric measurements, digital three-dimensional modelling, LiDAR sensor, 40 Velodyne VLP-16 41 1. Introduction 42 The use of remote sensing in agriculture is not new; it dates back decades. However, 43 recent technological advances in sensors, instrumentation, and wireless networks have brought 44 innovative levels of monitoring into raising livestock (Guo et al., 2017, 2019; Le Cozler et al., 45 2019). Sensors measuring animal welfare, health, growth rate, and physiological features, 46 including shape, size, and weight, either alone or in combinations, have been applied in animal 47 breeding. Kawasue et al. (2013) used three Kinect sensors to investigate the weight and size of 48 cattle, as well as their postures and shapes, with accuracies of up to 93% relative to manual 49 measurements. Other authors have also evaluated the body shape and temperature of black 50 cattle using a thermal camera and a Kinect sensor (Kawasue et al., 2017), the correlation 51 between manual measurements and stereo vision in horses (Pallottino et al., 2015), and the 52 3 combination of a binocular stereo vision system with the LabView development platform to 53 estimate pig body size and weight in indoor farm conditions (Shi et al., 2016). 54 The horse industry plays a part in national, state, and local economies in Europe, North 55 America, and Canada (Cross, 2019). The equine sector is diverse, involving agriculture, 56 business, recreation, racing, competitions, and the generation of specialized skills and general 57 employment across the board. Breeding, competition, and leisure activities involving horses are 58 important business concerns, and while horses are no longer used for primary transport, they 59 remain important assets. Although not currently popular in most countries, the consumption of 60 horse meat is increasing in several Western European countries due to its availability and 61 recognized nutritional value and mainly as an alternative to the traditional consumption of red 62 meat (Belaunzaran et al., 2015). 63 Equine production is one of the main agricultural activities in Spain, responsible for 0.5% of 64 the gross national product (Sánchez-Guerrero et al., 2018), with the Pura Raza Española horse 65 (PRE) being the most important breed in terms of historical census (328,706 horses in 65 66 countries) and economic impact on international trade (Sánchez-Guerrero et al., 2016). Typical 67 genetic selection programmes are based on functionality (dressage) and morphological 68 characteristics. Therefore, obtaining accurate zoometric measurements, characterized by high 69 repeatability and low variance among observers, can be considered ideal conditions in the 70 context of an animal breeding programme (Duensign et al., 2014). In the case of horses, the 71 close connection between biometric parameters, locomotion characteristics, and sports 72 performance is often the only tool for genetic improvement (Pallottino et al., 2015) and as an 73 indicator of functional longevity. In recent decades, indirect conformation assessment systems 74 based on linear morphological traits have been implemented for many equine breeds (Kuhnke 75 et al., 2019; Folla et al., 2019). This system has a series of advantages, as the scoring scale of 76 each region covers its biological range and a wide range of numerical classes can be used, 77 which allows subsequent statistical treatment of the data in a continuous scale. However, highly 78 4 trained qualifiers are required to apply the linear morphological system, and the relationship 79 between zoometric measurements and the linear scale used has not been established. The 80 main goal of the PRE breeding programme is to improve not only animal functionality but also 81 its conformation for sport performance (Sánchez-Guerrero et al., 2016). Therefore, in the PRE, 82 the importance of conformation has been shown in studies carried out to demonstrate its 83 relationship with dressage (Sánchez-Guerrero et al., 2017; Solé et al., 2013), functional 84 longevity (Solé et al., 2017), and genetic improvement of morphofunctional traits (Sánchez-85 Guerrero et al., 2017). 86 In the last decade, several computerized techniques have been proposed for the collection 87 of biometric data through images (Wu et al., 2004; Viazzi et al., 2014). Novel three-dimensional 88 systems can solve the problems posed by conventional two-dimensional vision systems, 89 including stereo photogrammetric techniques. However, these photogrammetric systems are 90 difficult to implement. 91 In recent years, new techniques that provide point clouds in a fast, non-invasive and 92 inexpensive way, such as the Kinect v2® (Microsoft Corp., Redmond, Washington) or Xtion Pro 93 (USUSTeK COMPUTER Inc., Taipei, Taiwan), as well as infrared (IR) light (Salau et al., 2017; 94 Kawasue et al., 2017; Viazzi et al., 2014; Mortensen et al., 2016; Guo et al., 2017; Pezzuolo et 95 al., 2018; Song et al., 2018; Guo et al., 2019), are being used to develop new method to 96 automate the collection of morphological information from livestock. In fact, Pezzuolo et al. 97 (2019) performed a metrological analysis of the Structure from Motion (SfM) approach, low-cost 98 LiDAR scanning, and the Microsoft Kinect v1 depth camera for three-dimensional measurement 99 of an animal's body, with specific reference to pigs. The results obtained demonstrated the high 100 potential of the 3D Kinect and a higher root mean square (RMS) value of LiDAR with respect to 101 Kinect and SfM, probably due to the collection approach based on individual profiles instead of 102 surfaces. 103 5 However, despite progress in the three-dimensional reconstruction of the bodies of animals, 104 no robust descriptor for the automatic estimation of conformation has currently been found in 105 equine species. Therefore, the main objective of this study is to explore a 3D reconstruction 106 approach for body measurements in horses using point cloud analysis. Our specific objectives 107 were (1) to develop a portable scanning system for domesticated animals; (2) to obtain valuable 108 zootechnical information through 3D point cloud analysis (that would avoiding wasting time and 109 money); and (3) to compare the accuracy between the digital and conventional manual 110 measurements on five PRE horses. 111 112 2. Materials and methods 113 2.1. Animals and morphological variables under study 114 Five PRE horses, belonging to the breeding programme of the “Yeguada Cartujana del 115 Hierro del Bocado” (Jerez de la Frontera, Cádiz, Spain), situated at latitude 36°41′ north and 116 longitude 06°09′ west, were analysed (Table 1). 117 Table 1. Ages and coat colours of the five Pura Raza Española horses used in this study. 118 Stallion 1 Stallion 2 Stallion 3 Mare 1 Mare 2 Age 5 years old 6 years old 8 years old 12 years old 18 years old Coat Colour Grey Bay Grey Grey Grey 119 The animals had been used for breeding and dressage exhibitions. To reduce possible 120 variations, the zoometric measurements taken by the manual method were always carried out 121 by the same veterinarian. For each horse, 17 zoometric measurements (15 linear and 2 122 angular) relating to sport performance (Sánchez-Guerrero et al., 2016, 2017) were collected 123 (Fig. 1). 124 6 Figure. 1. Graphical representation of the morphological measurements taken in the Pura Raza Espanola stallion 1 of the Yeguada Cartujana del Hierro del Bocado. Height at the withers (HW); Height to the lowest point at the withers (HWl); Height at the croup (HCr); Length of the body (LB); Length of the head (LH); Width of the head (WH); Length of the neck (LN); Length of the shoulder (LS); Length of the forearm (LFA); Dorso-sternum diameter (DSD); Length of the croup (LCr); Length of the femur (LF); Depth of the croup (DCr); Length of the gaskin (LG); Length of the buttock (LBu); Angle of the shoulder (AS); Angle of the croup (ACr). 125 The morphological measurements taken were as follows: 126 1. Height at the withers (HW): measured from the ground to the highest point of the 127 withers. 128 2. Height to the lowest point at the withers (HWl): measured from the ground to the lowest 129 point of the withers. 130 3. Height at the croup (HCr): measured from the ground to the highest point of the tuber 131 coxae. 132 4. Length of the body (LB): measured between the tubercle of the humerus and the ischial 133 tuberosity. 134 7 5. Length of the head (LH): distance from the nape to the cranial border of the snout. 135 6. Width of the head (WH): distance between the most protruding edge of the zygomatic 136 arches. 137 7. Length of the neck (LN): distance between the base of the ear and the middle point of 138 the spine of the scapula. 139 8. Length of the scapula (LS): distance between the wither and shoulder. 140 9. Length of the forearm (LFA): distance between the parallel straight lines drawn down 141 from the elbow and carpal joint midpoint. 142 10. Dorso-sternum diameter (DSD): distance measured from the lowest point in the wither 143 decline to the sternal area. 144 11. Length of the croup (LCr): distance between the coxal tuberosity at its midpoint and the 145 ischial tuberosity. 146 12. Length of the femur (LF): distance between the buttock and stifle 147 13. Depth of the croup (DCr): distance between the hip and stifle. 148 14. Length of the gaskin (LG): distance between the stifle and the hock 149 15. Length of the buttock (LBu): distance between the coxal tuberosity of the ilium and the 150 ischial tuberosity. 151 16. Angle of the shoulder (AS): angle formed by the line from the withers to the shoulder 152 with the horizontal. 153 17. Angle of the croup (ACr): angle formed by the line from the ischial tuberosity to the 154 tuber coxae with the horizontal. 155 156 2.2. Data acquisition and data processing 157 2.2.1 Conventional measurement system 158 8 For each horse, zoometric measurements were systematically collected using standard 159 measuring sticks, non-elastic measuring tape and zoometric compasses (Sánchez-Guerrero et 160 al., 2016). All measurements were taken from the left side of the horse while it was standing on 161 a hard surface and flat ground, assuming a natural position. The horses were positioned for 162 measurement with the front legs and hind feet parallel and as near to perpendicular as possible; 163 the toes were in line. No sedatives were used. 164 165 2.2.2. Measurements based on the point cloud system 166 The zoometric measurements were acquired through an optical remote sensing technique 167 using a LiDAR sensor. The sensor uses laser light to obtain a dense sampling of the target, 168 producing accurate measurements in three dimensions. The reflection of laser light off the target 169 is detected and analysed by the receivers in the LiDAR sensor. These receivers, consisting of a 170 receiver telescope, filters, and an optical detector, record the precise time between when the 171 laser pulse left the system and when it returned to calculate the limit distance between the 172 sensor and the target. These spatially organized post-processed LiDAR data are known as point 173 cloud data, which is a large data set consisting of 3D point data. Each LiDAR point can have an 174 assigned classification defining the type of object that reflected the laser pulse. 175 The LiDAR used was a Velodyne VLP-16 (Velodyne Lidar, Inc., California, EEUU), which is 176 a high-precision 3D laser sensor LiDAR with a rotating head containing a certain number of 177 semiconductor lasers or laser diodes. Each of the lasers has its own detector (Fig. 2). 178 9 Figure 2. Representation of the lasers of the Velodyne VLP-16 179 The VLP-16 sensor measures the reflectivity of an object with 256-bit resolution 180 independent of laser power and distance within an interval of 1-100 m. Commercially available 181 reflectivity standards were used for absolute reflectivity calibration, which are stored in a 182 calibration table within the field-programmable gate array (FPGA) of the VLP-16. 183 The VLP-16 scanner has 16 individual lasers/detectors arranged in a 30° FOV, which yields 184 a vertical resolution of 2.0°. This sensor has an FOV symmetrical with the respect to the 185 horizontal plane, and points can be obtained up to 100 m away at a rate of approximately 186 300,000 per second in single return and 600,000 for dual return. The horizontal FOV is 360°, 187 with an adjustable rotation frequency between 5 and 20 Hz. 188 Thanks to the divergence of the laser beam, a single shot can hit multiple objects, and 189 different returns will occur. The VLP-16 has the ability to analsze the multiple returns and report 190 the strongest return, the last return, or both (dual return or dual mode). Multiple returns occur 191 when a laser pulse strikes the horse in a location that does not completely block the path of the 192 pulse, allowing the remaining portion of the pulse to continue to the next seen object. 193 Communication between the sensor and the computer performing the data analysis was 194 performed was carried out through an interface box with an Ethernet cables and user-assigned 195 IP addresses. 196 16 Table 2. Zoometric measurements, relative error, v2 error and Pearson correlations in five Pura Raza Española horses from Estrepe Cartujana 286 obtained using the conventional procedure (manual) and the LiDAR system (sensor) 287 M Stallions 1 Stallions 2 Stallions 3 Mare 1 Mare 2 Global v2 r m s r.e. m s r.e. m s r.e. m s r.e. m s r.e. HW 156 160 -2.56 159 149 6.29 164 156 4.88 160 149 6.88 163 150 7.98 1.14 -0.32 HWl 151 141 6.62 150 148 1.33 153 155 -1.31 149 146 2.01 153 150 1.96 3.59 0.60 HCr 158 145 8.23 159 155 2.52 159 140 11.95 160 154 3.75 162 161 0.62 23.17 0.77 LB 156 150 3.85 155 155 0.00 159 153 3.77 159 158 0.63 170 156 8.24 0.86 0.39 LH 60 63 -5.00 58 57 1.72 59 62 -5.08 61 57 6.56 62 60 3.23 0.52 0.06 WH 23 29 -26.09 23 22 4.35 24 23 4.17 21 17 19.05 23 25 -8.70 3.92 0.64 LN 74 78 -5.41 73 70 4.11 79 73 7.59 78 78 0.00 78 75 3.85 0.64 0.25 LS 62 53 14.52 62 54 12.9 62 67 -8,06 66 73 10.61 69 70 -1.45 0.98 0.71 LFA 54 54 0.00 47 56 19.15 49 40 18.37 44 37 15,91 47 51 8.51 1.07 0.48 DSD 72 54 25.00 71 54 23.94 74 62 16.22 74 66 10.81 77 67 12.99 3.58 0.90 LCr 56 50 10.71 54 54 0.00 52 51 1.92 52 54 3.85 53 55 3.77 1.86 -0.45 LF 53 44 16.98 50 42 16.00 51 48 5.88 48 41 14.58 45 42 6.67 0.86 0.53 DCR 52 48 7.69 49 39 20.41 52 53 1.92 52 35 32.69 53 49 7.55 17.70 0.47 LG 51 64 25.49 48 49 2.08 50 50 0 57 62 8.77 54 55 1.85 2.32 0.62 LBu 47 45 4.26 45 41 8.89 48 47 2.08 42 39 7.14 45 47 4.44 0.25 0.79 AS 57 66 15.79 57 86 50.88 56 62 10.71 58 83 43.1 57 65 14.04 230.60 0.66 ACr 14 25 78.57 18 10 44.44 18 23 27.78 17 21 23.53 18 20 11.11 4.17 -0.55 Measurements (M); conventional procedure (m) and the LiDAR system (s); relative error (r.e.); Pearson correlations coefficient (r); MHeight at withers1 (HW); 288 Height at croup1 (HCr); Length of head1 (LH); Width of head1 (WH); Length of neck1 (LN); Length of shoulder1 (LS); Length of forearm1 (LFA); Length of body1 289 (LB); Length of croup1 (LCr); Length of femur1 (LF); Length of gaskin1 (LG); Depth of croup1 (DCr); Length of buttock1 (LBu); Dorso-sternum diameter1 (DSD); 290 Angle of shoulder2 (AS); Angle of croup2 (ACr). 1Centimters 2Degrees 291 17 Size and body conformation are critically important traits in nearly all horses breeds, which 292 are presumably subject to a strict process of selection over time (Brooks et al., 2010). The 293 zoometric measurements recorded during this study were similar to those obtained previously in 294 the same breed (Molina et al., 1999; Gómez et al., 2009). Although the bay PRE horses are the 295 tallest in this study, the height of stallion 2 was between that of stallions 1 and 3. 296 The measurements were taken both in mares and stallions to measure the possible 297 difference in accuracy of the measurement at the horse's mane; in mares, the mane is usually 298 cut, and in stallions it is usually left long (Fig. 1 and 4). In our results, stallions had a higher 299 relative error than mares (10.83 and 8.80, respectively) and a higher average v2 value. 300 Among the three stallions, the relative error ranged between 0 (HC, LFA, LCr and LC) and 301 78.57 (Acr). A total of 7.84% of the zoometric measurements studied had a relative error of 0; 302 for 52.94% had an error > 0 and <10; for 21.57% had an error ≥10 and <20; for 11.76% had an 303 error ≥20 and <30; and for 5.88% had an error ≥ 30. For the two mares studied, the relative 304 error ranged between 0 (LN) and 43.10 (AS). A total of 2.94% of the zoometric measurements 305 studied had a relative error of 0; for 64.71% had an error > 0 and <10; for 23.53% had an error 306 ≥10 and <20; for 2.94% had an error ≥20 and <30; and for 5.88% had an error ≥ 30 (Table 2). 307 The mean relative error was 10.83% and 8.00% without angle measurements. A total of 82.5% 308 of the studied features had an accuracy (v2) of less than 5%; only one of the angle 309 meaurements (AS) had a higher v2 value. The Pearson correlation between the manual and 310 sensor measurements ranged from 0.90 (DSD) to 311 -0.55 (Acr) or -0.45 (LCr) if the angles were no taken into account (Table 3). The angles are, in 312 general, very difficult to measure also in the traditional system and in the linear morphological 313 assesments since they are influenced by the horse's posture (Sánchez et al., 2013). We 314 suggest that perhaps this problem could be solved with the use of LiDAR or a similar technique, 315 since the scan could be performed when the animal is perfectly poised, and in real life, the 316 horse is always moving to some degree while zoometric measurements are being taken. The R² 317 18 and RMSE estimated by comparing the zoometric measurements obtained using the 318 conventional procedure (manual) and the LiDAR system (sensor) were 0.97 and 7.62, 319 respectively (Figure 8). 320 y = 2.4251 + 0.94268 * x r = 0.9861; R 2 =0.9723; RMSE=7.6172 020 40 60 80 100 120 140 160 180 Manual zoometric measurement 0 20 40 60 80 100 120 140 160 180 LIDAR zoometric measurement 0.95 Confidence Interval 321 Figure 8. A scatter plot with regression analysis between the coefficient of determination (R²) and root-322 mean-square error (RMSE) comparing zoometric measurements obtained for five Pura Raza Española 323 horses from Estrepe Cartujana performed using the conventional procedure (manual) and the LiDAR 324 system (sensor) 325 326 Thus, our results differ from those reported for different body measures on Mediterranean 327 buffaloes (Negretti et al., 2007), correlating traditional measures with predicted ones using 328 image analysis and obtaining coeficients ranging between 0.91 and 0.99 depending on the 329 measured trait. Moreover, Pallottino et al. (2015) correlated traditional measurements with those 330 obtained with a dual web-camera system (r = 0.998) in Lipizzan horses. In any case, It is 331 19 important to highlight that the Pearson correlation coefficients are not the best way to detect the 332 differences between two methods of measurement, since we are not as interested in whether 333 they increase or decrease in the same direction. The main interest is to ensure that the 334 difference between the two measures is as small as possible. This could be better estimated 335 with the relative error. For the body size measurements of sheep, a low-cost dual web camera 336 was used, yielding a mean relative error of 5% (Menesatti et al., 2014). In addition, 16 adult 337 dairy cows were studied with pairs of pictures, taken with a portable instrument incorporating 338 two synchronized cameras, with photogrammetric measurements having an accuracy within 0.5 339 cm (Gaudioso et al., 2014). The use of a structured light depth camera for three-dimensional 340 body measurements of dairy cows in free-stall barns has also been evaluated (Pezzuolo et al., 341 2018), obtaining coefficients of determination of R2 > 0.84 and deviations lower than 6% with 342 respect to manual measurements. As in our case, lower performances were obtained for the 343 angle measurements (back slope: R2 = 0.12). In our study, the AS measurement had a very 344 high v2 value, while ACr had a low v2 value in both genders. This is the first study in which 17 345 zoometric measurements were analysed, with the majority of the previous studies having fewer 346 traits analysed (ranging from 3 to 8 body measurements (Huang et al., 2018; Pezzuolo et al., 347 2018)). Taking accurate manual measurement data as validation criteria, LiDAR methodology 348 was used in another study to measure three live cattle, and the experimental results showed 349 that the final deviations were close to 2 mm and within approximately 2% (Huang et al., 2018). 350 The reproducibility of these traits in the linear morphological system ranged from 0.86 to 0.99 in 351 PRE (Sánchez et al., 2013). In studying another group of zoometric measurements, Gaudioso 352 et al. (2014) found a standard deviation of 2.0 cm. It shold be noted that we have not been able 353 to study the precision of these 17 zoomometric measurements since replications were not 354 carried out. 355 Once the methodology proposed in this project has been presented and the results 356 obtained from this process have been analysed, the final development of two conclusions can 357 20 be presented, one from a generic point of view and another from a specific point of view, in 358 addition to a roadmap for future developments. 359 4. Conclusions 360 A general methodology has been created that establishes a data evaluation process, so that in 361 the future, applications can be developed in which we can encompass the entire process. The 362 use of a LiDAR system has simplified the taking of measurements, giving rise to a method that 363 can obtain optimal results in monitoring the growth of the livestock. This methodology specifies 364 a clear, precise and logical process. 365 In theory, this method can be extrapolated to any equine breed, as well as to other livestock 366 species. Although it is a pilot study and its accuracy must be improved, it could help in 367 preventing excess spending and stressing animals with the current manual zoometric 368 measurement. More specificically, the following conclusions can be drawn based on the results 369 obtained from implementing the methodology. The approach carried out in this work verifies the 370 viability of the noncontact measurement of large livestock. Knowledge of the scalable growth 371 and animal welfare at the location where the animals are housed will improve the quality of the 372 animals and genetic breeding. However, as can be seen in the results section, due to the 373 constant movement of the horses and the limitation of single-frame analysis, it is not possible to 374 fully analyse the three-dimensional image of the horse. The geometric accuracy of these 375 sensors is closely related to the quality of their external orientation. This external orientation is 376 what we will obtain from the integrated navigation systems (GNSS and INS). The laser scanner 377 measures only the vector oriented from the opening of the laser system to an object point 378 normally on the ground, the horse. Three-dimensional points can only be calculated if at any 379 moment the position and orientation of the LiDAR sensor with respect to a coordinate system is 380 known. A future study will try to improve these drawbacks by integrating two synchronized point 381 21 clouds through two LiDARs positioned on each side of the animal and performing repeated 382 measurements with the conventional procedure (manual) and the LiDAR system (sensor). 383 For this problem, in the near future this system will continue to be optimized by activating the 384 LiDAR sensor and an integrated navigation system that will allow the sensor to move so that the 385 horse can be rotated while it is being scanned. This will further allow a complete reconstruction 386 of the horse in 3D for greater accuracy in obtaining the zoometric variables. 387 388 Acknowledgements 389 The research was supported by project AGL2016-78964-R funded by the Spanish Ministry of 390 Economics and Competence. Additionally, the authors want to thank the Yeguada Cartujana-391 Hierro del Bocado in Jerez (Cádiz) for providing us with their facilities and access to their 392 animals. 393 References 394 Belaunzaran, X., Bessa, R.J.B., Lavín, P., Mantecón, A.R., Kramer, J.K.G., Aldai, N., 2015. 395 Horse-meat for human consumption - Current research and future opportunities (Review). 396 Meat Science. 108, pp. 74-81. https://doi.org/10.1016/j.meatsci.2015.05.006 397 Brooks, S.A., Makvandi-Nejad, S., Chu, E., Allen, J.J., Streeter, C., Gu, E., McCleery, B., 398 Murphy, B.A., Bellone, R., Sutter, N.B., 2010. Morphological variation in the horse: 399 defining complex traits of body size and shape. Animal Genetics. 41(2), pp.159–65. 400 https://doi.org/10.1111/j.1365-2052.2010.02127.x 401 Cross, P., 2019. Global Horse statistics internal. Global Horse statistics. HPAB (HiPoint Agro 402 Bedding Corp.) Helping the Planet through Animal Biodiversity. Global Horse. 403 Duensing, J., Stock, K.F., Krieter, J., 2014. Implementation and prospects of linear profiling in 404 the warmblood horse. Journal of Equine Veterinary Science. 34, pp. 360–368. 405 https://doi.org/10.1016/j.jevs.2013.09.002 406 Folla, F., Sartori, C., Guzzo, N., Pigozzi, G., Mantovani, R., 2019. Genetics of linear type traits 407 scored on young foals belonging to the Italian Heavy Draught Horse breed. Livestock 408 Science. 219, pp. 91–96. https://doi.org/10.1016/j.livsci.2018.11.019 409 Gaudioso, V., Sanz-Ablanedo, E., Lomillos, J.M., Alonso, E., Javares-Morillo, L., Rodríguez, P., 410 2014. “Photozoometer”: A new photogrammetric system for obtaining morphometric 411 22 measurements of elusive animals. Livestock Science.165, pp 147–156. 412 https://doi.org/10.1016/j.livsci.2014.03.028 413 Gómez, M.D., Goyache, F., Molina, A., Valera, M., 2009. Sire × stud interaction for body 414 measurement traits in Spanish Purebred horses. Journal of Animal Science. 87, pp. 2502–415 2509. https://doi.org/10.2527/jas.2008-0841 416 Guo, H., Ma, X., Ma, Q., Wang, K., Su, W., Zhu, D., 2017. LSSA_CAU: An interactive 3d point 417 clouds analysis software for body measurement of livestock with similar forms of cows or 418 pigs. Computers and Electronics in Agriculture. 138, pp. 60–68. 419 https://doi.org/10.1016/j.compag.2017.04.014 420 Guo, H., Bo Li, Z., Ma, Q., Zhu, D., Su, W., Wang, K., Marinello, F., 2019. A bilateral symmetry 421 based pose normalization framework applied to livestock body measurement in point 422 clouds. Computers and Electronics in Agriculture. 160, pp. 59-70. https://doi.org/ 423 10.1016/j.compag.2019.03.010 424 Huang, L., Li, S., Zhu, A., Fan, X., Zhang, C., Wang, H., 2018. Non-contact body measurement 425 for qinchuan cattle with LiDAR sensor. Sensors. 18(9), pp. 3014. 426 https://doi.org/10.3390/s18093014 427 Kawasue, K., Ikeda, T., Tokunaga, T., Harada, H., 2013. Three-dimensional shape 428 measurement system for black cattle using KINECT sensor. International Journal of 429 Circuits, System and Signal Processing. 7, pp. 222–230. 430 Kawasue, K., Win, K.D., Yoshida, K., Tokunaga, T., 2017. Black cattle body shape and 431 temperature measurement using thermography and KINECT sensor. Artificial Life and 432 Robotics. 22, pp. 464–470. https://doi.org/10.1007/s10015-017-0373-2 433 Kuhnke, S., Bär, K., Bosch, P., Rensing, M., Borstel, U.K.V., 2019. Evaluation of a Novel 434 System for Linear Conformation, Gait, and Personality Trait Scoring and Automatic 435 Ranking of Horses at Breed Shows: A Pilot Study in American Quarter Horses. Journal of 436 Equine Veterinary Science. 78, pp. 53–59. https://doi.org/10.1016/j.jevs.2019.04.002 437 Le Cozler, Y., Allain, C., Caillot, A., Delouard, J.M., Delattre, L., Luginbuhl, T., Faverdin, P., 438 2019. High-precision scanning system for complete 3D cow body shape imaging and 439 analysis of morphological traits. Computers and Electronics in Agriculture. 157, pp. 447–440 453. https://doi.org/10.1016/j.compag.2019.01.019 441 Menesatti, P., Costa, C., Antonucci, F., Steri, R., Pallottino, F., Catillo, G., 2014. A low-cost 442 stereovision system to estimate size and weight of live sheep. Computers and Electronics 443 in Agriculture. 103, pp. 33–38. https://doi.org/10.1016/j.compag.2014.01.018 444 Molina, A., Valera, M., Dos Santos, R., Rodero, A., 1999. Genetic parameters of 445 morphofunctional traits in Andalusian horse. Livestock Production Science. 60, pp. 295–446 303. https://doi.org/10.1016/S0301-6226(99)00101-3 447 Mortensen, A.K., Lisouski, P., Ahrendt, P., 2016. Weight prediction of broiler chickens using 3d 448 computer vision. Computers and Electronics in Agriculutre. 123, pp. 319–326. 449 https://doi.org/10.1016/j.compag.2016.03.011 450 Negretti, P., Bianconi, G., Bartocci, S., Terramoccia, S., Verna, M., 2008. Determination of live 451 weight and body condition score in lactating Mediterranean buffalo by Visual Image 452 Analysis. Livestock Science. 113, pp. 1–7. https://doi.org/10.1016/j.livsci.2007.05.018 453 Pallottino, F., Steri, R., Menesatti, P., Antonucci, F., Costa, C., Figorilli, S., Catillo, G., 2015. 454 Comparison between manual and stereovision body traits measurements of Lipizzan 455 23 horses. Computers and Electronics in Agriculture. 118, pp. 408–413. 456 https://doi.org/10.1016/j.compag.2015.09.019 457 Pezzuolo, A., Guarino, M., Sartori, L., Marinello, F., 2018. A Feasibility study on the use of a 458 structured light depth-camera for three-dimensional body measurements of dairy cows in 459 free-stall barns. Sensors. 18 (2), pp. 673. https://doi.org/10.3390/s18020673 460 Preisinger, R., Wilkens, J., Kalm, E., 1991. Estimation of genetic parameters and breeding 461 values for conformation traits for foals and mares in the Trakehner population and their 462 practical implications. Livestock Production Science. 29 (1), pp. 77-86. 463 https://doi.org/10.1016/0301-6226(91)90121-6 464 Poly, J., Poutous, M., Calomiti, S., Suzanne C., 1967. Le controle laitier mensual alterné (AT). 465 I._Précision vis-a-vis d’un controle mensual ou bimestriel pour la produciton “de lait en 466 305 jours”. Ann. Zootech. 16, pp. 183–190. 467 Sánchez, M.J., Gómez, M.D., Molina, A., Valera, M., 2013. Genetic analyses for linear 468 conformation traits in Pura Raza Español horses. Livestock Science. 157, pp. 57–64. 469 https://doi.org/10.1016/j.livsci.2013.07.010 470 Sánchez-Guerrero, M.J., Molina, A., Gómez, M.D., Peña, F., Valera, M., 2016. Relationship 471 between morphology and performance: Signature of mass-selection in Pura Raza Español 472 horse. Livestock Science. 185, pp.148–155. https://doi.org/10.1016/j.livsci.2016.01.003 473 Sánchez-Guerrero, M.J., Cervantes, I., Molina, A., Gutiérrez, J.P., Valera, M., 2017. Designing 474 an early selection morphological linear traits index for dressage in the Pura Raza Español 475 horse. Animal. 11(6), pp. 948-957. https://doi.org/10.1017/S1751731116002214 476 Sánchez-Guerrero, M.J., Negro-Rama, S., Daymas-Peyrás, S., Solé-Berga, M., Azor-Ortiz, P.J., 477 Valera-Córdoba, M., 2018. Morphological and genetic diversity of Pura Raza Español 478 horse regarding to the coat colour. Animal Science Journal. 90, pp.14–22. 479 https://doi.org/10.1111/asj.13102 480 Salau, J., Haas, J.H., Junge, W., Thaller, G., 2017. A multi-kinect cow scanning system: 481 calculating linear traits from manually marked recordings of holstein-friesian dairy cows. 482 Biosystem Engineering. 157, pp. 92–98. 483 https://doi.org/10.1016/j.biosystemseng.2017.03.001 484 Shi, C., Teng, G., Li, Z., 2016. An approach of pig weight estimation using binocular stereo 485 system based on LabVIEW. Computers and Electronics in Agriculture. 129, pp. 37–43. 486 https://doi.org/10.1016/j.compag.2016.08.012 487 Solé, M., Santos, R., Gómez, M.D., Galisteo, A.M., Valera, M., 2013. Evaluation of conformation 488 against traits associated with dressage ability in unridden Iberian horses at the trot. 489 Research in Veterinary Science. 95, pp. 660–666. 490 https://doi.org/10.1016/j.rvsc.2013.06.017 491 Solé, M., Sánchez, M.J., Valera, M., Molina, A., Azor, P.J., Sölkner, J., Mészáros, G., 2017. 492 Assessment of sportive longevity in Pura Raza Español dressage horses. Livestock 493 Science. 203, pp. 69-75. https://doi.org/10.1016/j.livsci.2017.07.007 494 Song, X., Bokkers, E.A.M., van der Tol, P.P.J., Koerkamp, P.W.G.G., van Mourik, S., 2018. 495 Automated body weight prediction of dairy cows using 3-dimensional vision. Journal of 496 Dairy Science.101 (5), pp. 4448–4459. https://doi.org/10.3168/jds.2017-13094 497 Viazzi, S., Ismayilova, G., Oczak, M., Sonoda, L.T., Fels, M., Guarino, M., Vranken, E., Hartung, 498 J., Bahr, C., Berckmans, D., 2014. Image feature extraction for classification of aggressive 499 24 interactions among pigs. Computers and Electronics in Agriculture. 104, pp. 57–62. 500 https://doi.org/10.1016/j.compag.2014.03.010 501 Wu, J., Tillett, R., McFarlane, N., Ju, X., Siebert, J.P., Schofield, P., 2004. Extracting the three-502 dimensional shape of live pigs using stereo photogrammetry. Computers and Electronics 503 in Agriculture. 44, pp. 203–222. https://doi.org/10.1016/j.compag.2004.05.003 504 505 506