We have developed a growth chamber to study growth in a 24 h continuous fashion. It is constructed of removable panels of aluminium integrating an artificial vision system. The corresponding vision system comprises an image capture device, an illumination system and an image-processing unit. We designed the growth chamber with a reconfigurable structure to accommodate plants of different sizes. It has an inner volume of 300 × 300 × 700 mm, and an external size of 800 × 900 × 700 mm.

The aluminium profiles used for the construction allow the reorganization of the light and cameras, allowing a high versatility to perform experiments with plants of different sizes. Figure 1 shows a schematic representation of the chamber with all the systems that integrate the prototype.

The study of circadian events requires an unusual capacity to fetch variations of form and/or shape at time intervals that can be as low as minutes. In practical terms, both nutation and growth have to be separated from each other in order to quantify both parameters. We designed acomputer vision system to analyze the behaviour of circadian rhythm of one or more plants in a continuous way. In order to achieve that, we used two cameras and a configurable software tool, which allows measuring the morphological characteristics during the programmed periods of day/night. The vision system is divided in three main elements as described below.

We used three types of plants: the ornamentals Petunia x hybrida and Antirrhinum majus line 165E and the cactus Opuntia ficus-indica. We grew the double haploid Mitchell line W115 from petunia and Antirrhinum majus line 165E using standard greenhouse conditions as described previously [29,30]. Plants grown under natural conditions were transferred to the developed growth chamber that had been set up to respect the day-night rhythm of the year time. Nevertheless, plants were grown inside the chamber in order to achieve perfect synchronization with the artificial day-night cycling for 6 days before sampling started.

Antirrhinum plants corresponded to 165E wild type plants and the mutant nana [31], identified as a circadian clock mutant in transcriptomic experiments (Egea-Cortines et al., unpublished observations).

Cladodes of the Opuntia ficus-indica var. Ofer [32] were planted in pots and grown as described [33] under natural conditions in the greenhouse. Plants were brought to the growth chamber and left for 6 days before sampling started.

In order to select the organs for automatic sampling of morphological characters allowing growth analysis we established three criteria:

Organ orientation from the camera. We selected organs that display a lateral profile during the longest period of capture.

Preliminary experiments performed to set up the system described in this work showed that in many cases plant organs grew out of the range of the Region of Interest (ROI) originally set. Indeed, stems from wild-type and nana mutants of snapdragon consistently grew out of the ROI as a result of sheer growth, nutation and in some cases organ deterioration with aging. Thus we had to develop a system that would recalculate the ROI position and size. Plant organs showed different growth and mobility patterns as we could ascertain in the current work. Both parameters changed with differing speed. Furthermore, we could discriminate between slow movements registered for growth and stem circumnutation and fast movements identified in flower growth and opening.

In order to measure a single organ, the Automated Image Analysis module allows selecting manually a ROI where a set of computer vision methods will be applied. Under slow movements, selection of a ROI was in most cases enough to measure. However, during rapid movements like flowering, it was necessary to change the ROI shape.

The algorithm to obtain an Adaptive ROI (AROI) was developed in such a way that it calculated the rectangle circumscribing the organ studied, and maintained a constant distance between the rectangle and the ROI. It comprises the following steps:

In the first image, the user selects manually a ROI. Within the ROI, a blobcorresponding to the organ studied is determined.

Figure 10(a,b,c) show a sequence of three images of Antirrhinum flowers taken 24 h apart where the originally fixed ROI is not capable of obtaining the image characteristics as growth has displaced its position. Figure 10(d,e,f) show the same sequence, where the displacement and size increment of the ROI allow the correct acquisition of the object parameters.

Using two independent illumination systems for day and night and two cameras for each period allows a better adjustment than those systems based on a single illumination and camera combination [16]. It allows an exact adjustment of the vision system parameters i.e., optic opening, exposure times and camera adjustments for every period. As a result better images with higher contrast can be obtained in each period, facilitating the segmentation process and follow up of the plant organs. Furthermore, it reduces the complexity of the algorithms used and a higher resolution can be achieved for morphological studies. Figure 11 shows how samples are acquired in a sequence (nigh1-day1-night2-day2 …) generating two types of transitions: night-day (T^nd) and day-night (T^dn). This process has as major disadvantage the abrupt transition between grey levels between the day/night shift periods (see Figure 12(a)). We solved this problem by using Equations (4) and (5) to correct the values of variables based on pixel counting like area, length, centre of gravity and perimeter.

Being V_j a variable depending on the pixels counting of the image at instant j, the corrected value of the variable due to a transition from a night period (j = 1, …, n) till a day period (j = n + 1, … n + d) is obtained using Equation (4), while the corrected value of the variable due to a transition from a day period (j = n + 1, …, n + d) till a night period (j = n + d + 1, … 2n + d) is obtained using Equation (5). Being n the number of measures along a night period and being d the number of measures along a day period: (4) V j n d = V j + K 1 + K 2 , j = 1 , … , n , n + 1 , … , n + d K 1 = V n + 1 − V n K 2 = 1 n − 1 ∑ j = 1 n − 1 V j + 1 − V j (5) V j d n = V j + K 3 + K 4 , j = n + 1 , … , n + d , n + d + 1 , … , 2 n + d K 3 = V n + d + 1 − V n + d K 4 = 1 d − 1 ∑ j = n + d + 1 2 n + d − 1 V j + 1 − V j

Figure 12(b) shows the results obtained after applying the correction algorithm on the Y displacement of the centre of gravity of an Opuntia cladode for a complete experiment comprising 7 days at an acquisition rate of 60 images per hour.

As we can observe, the correction suppresses the sharp changes due to illumination transitions, giving a continuous value of displacement on the gravity centre in the Y axis (GCY).

Studies usingArabidopsis plants have shown that hypocotyl elongation occurs during the night period [34], and leaves of Populus tobacco and Arabidopsis maintain a near-dusk diel growth pattern even in in vitro conditions [19,35]. We used the double haploid cultivar of Petunia x hybrida Mitchell as it is highly amenable to transformation and has been extensively used to study flower development [36]. The petunia flower in the Mitchell cultivar highly resembles the P.axillaris flower as it has a long tube and a large limb. We used two independent plants to monitor growth of several flowers. However, measuring the development of the flower required the decomposition of growth into:

the nutation angle of the flower referred to the horizontal axis

We measured the nutation angle of the Petunia flowers during development. We observed larger changes in the Y axis as would be expected. Interestingly flowers showed a single pulse of movement shortly after dawn and dusk visible as changes in angle referred to the horizontal axis (Figure 13(a)). Although there were marked differences between flowers, during early stages of development, they tended to maintain a roughly similar angle of circa 50° during the day. At later stages of development, different flowers tended to adopt a final angle that could differ from each other by as much as 40°.

Petal growth is comprised of two periods, an initial period where cell division is responsible of most of the growth and a second period that is due to cell expansion [37]. We observed an almost linear increase in floral length during a period of two days before flower opening. The third day coincided with a sharp increase in floral length that almost doubled its size in a time span of less than 4 h (Figure 13(c)) coinciding with the measurements of floral length, total floral area showed an increase at the beginning of day 3 (TZ2-6; Figure 13(d)).

We calculated the velocity of growth for time zero till time 72 h Figure 13(b). Although total floral length and area do not decrease throughout the stages analyzed, the corresponding relative growth speed was sometimes negative indicating that the flowers had modified their position towards the camera.

The motion vector—a measurement of nutation—showed a sharp change shortly after dawn (Figure 14). As flowers matured and the limb expanded to its full size, the motion vector showed a lower activity. We calculated the absolute movement covered by the flowers as a function of age and found that throughout development there was a quick movement of the GC of the flower shortly after dawn (Figure 14). In contrast to floral length and area that showed a single strong increase in size at dawn and early morning, the MV of the flowers presented a progressive dampening, as young flowers had larger changes than mature flowers. All above measures have been obtained with a resolution of one image every 10 min for a 3 days period.

Automatic sampling based on computer vision systems is an important asset as the number of samples taken allow a finer dissection of the process and avoids aliasing, i.e., the imputation of a lack of rhythm or false negative identification of differences as a result of samples being taken too wide apart. The data obtained with the aforementioned computer vision system shows that contrary to the expected growth of leaves that occur during the night, the largest change in petal size occurs in a very short period and happens at dawn and early morning. This is an interesting feature of the Petunia petal as scent production in Petunia is circadian regulated and happens at night [38]. Our experimental prototype also allows us to develop further tools to study effects of photoperiod on floral development, an important aspect for the ornamental industry.

In order to test our system with a plant of a different structure, we analysed growth of cladodes of the prickly pear Opuntia ficus-indica. We introduced plants in the growth chamber for a period of 7 days with a photoperiod of 16 h light and 8 h dark. The measurement resolution was of one image every 10 min and the amount of data gathered was 1,008 per variable measured.

Converting the data obtained to accumulated area and length (Figure 15) showed that indeed Opuntia cladodes show a rhythmic pattern, however the growth pattern did not adjust to a diel pattern. In fact we uncovered that the accumulated area and length did not directly match as there were certain measured times that registered growth in length but not in area (Table 3). This decoupling of proximo-distal versus lateral axis growth coincides with the current genetic model of leaf area as two independent sets of genes controlling these processes. However the cladode is not a true leaf, but a modified stem. Thus the data obtained with the system developed allows the development of testable hypothesis about the genetic and environmental control of growth in cladodes that should share components with the leaves. Looking at the compactness of the cladodes (Figure 16) we also could confirm the previous assumption as during growth, compactness departed from a perfect circle that would be represented by a value of 1.