Sample rate
The sample rate is the number of images EthoVision XT analyzes per second. Selecting a certain sample rate does not mean that the program can always analyze data at that rate. If the computer processor load is too high, EthoVision XT may skip a sample and analyze the next one. Skipped samples result in missed samples (see below).
The maximum sample rate is the camera frame rate. The sample rate you set in EthoVision XT can only be the frame rate divided by an integer. For example, for PAL video it is 25, 12.5, 8.33, etc.
important Some digital cameras support very high frame rates. However, subject detection at such rates requires a lot of processor capacity. To prevent that EthoVision XT discards samples while tracking live, do not set the frame rate and sample rate too high. Check the percentage of missed samples (see View the detection statistics) to make sure the EthoVision XT can handle the selected frame rate.
What is the optimal sample rate?
Setting the correct sample rate is very important. If the rate is too high, the noise caused by small movements of your animal will be picked up and give an overestimate of dependent variables such as the distance moved. If the sample rate is too low, you will loose data and get an underestimate of the distance moved. See also the picture below.
General guidelines for the sample rate
We strongly recommend that you determine the optimum sample rate for your specific setup and animals (see To determine the optimal sample rate). If, for instance, your treatment causes hyperactivity, you will need a higher sample rate for hyperactive animals than somnolent animals.
note In the list below, the sample rate for NTSC video (29.97/s) is approximated to 30/s.
Rodents
▪Rat, single subject, center-point only: 5/s
▪Mouse, single subject, center-point only: 12/s
▪Rat/Mouse Behavior recognition: 25-31/s (depending on TV standard)
▪Rat/Mouse, three body points: 25/s (PAL) to 30/s (NTSC or digital camera)
▪Rat/Mouse, three body points combined with marker-assisted identification: 12.5/s (PAL) to 15/s (NTSC or digital camera)
▪Rat/Mouse, three body points combined with Deep learning-based identification: 25/s or higher. See Deep learning: Requirements
Other mammals
▪Tree shrew (Tupaia): 6-12/s
Fish
▪Zebrafish, adult: 5-6/s
▪Zebrafish larva: For analog camera: 25/s (PAL) or 30/s (NTSC). For digital camera: 30/s or 60/s*
(*) For rapid movements you may want to track with a higher sample rate. It depends on the number of tracked subjects, the video resolution, the camera settings and the processor speed of your computer whether that is possible.
▪Damselfish: 5/s
▪Goldfish: 0.5/s
Arthropods
▪Mite: 1/s
▪Parasitic wasp: 2/s
▪Tick: 3/s
To determine the optimal sample rate
The optimal sample rate is the minimum sample rate that provides an accurate estimation of the dependent variables (distance, velocity, etc.) without including the redundant information due to phenomena other than the 'real' locomotion. For example, for an animal walking in a straight line the data points will never be in a straight line because the center-point of the subject shifts laterally with each step. In order to distinguish between 'real' movement and effects like the one described above, you can calculate dependent variables like distance moved using the maximum or a lower sample rate.
1.Create new Detection Settings and specify the maximum sample rate. Be aware that the maximum sample rate also depends on the performance of your computer, the number of animals you track, and the video resolution.
2.Acquire data with those Detection Settings.
3.Calculate the dependent variable you are interested in.
4.Export the data to Excel and plot the dependent variable values against the sample rate. In the example below, distance moved is used.
5.Repeat the steps above by selecting smaller sample rates.
Once the data are plotted as in the figure below, there should be a range of sample rates for which the dependent variable value does not change much (plateau). This means that slight changes in the sample rate do no result in loss of information, or addition of redundant information (noise and movements like body wobble).
Low sample rates result in loss of useful information, because the sinuosity of the original path is removed. Therefore, the total distance moved is usually decreased.
High sample rates result in acquisition of redundant information. In the case of body wobbling, and assuming that the animal is moving along a straight line, the lateral shift of the body center causes the total distance moved to be longer than the 'real' one. With Track smoothing you can filter out 'noise' as a result of body wobble. See Smooth the Tracks
The actual sample rate may be lower than the maximum you set, because an image cannot be captured until the previous one is processed. If the sample rate you define is too high, EthoVision will miss samples (up to 1% is acceptable) and the processor load will be high. The percentage of missed samples is shown in the Detection performance pane and in the Trial List as a System Variable. You can calculate the number of missed samples in acquired tracks with the Number statistic of continuous variables (e.g., velocity). If your processor load is larger than 100, and there are large amounts of missed values, you will have to lower the sample rate.
tip After acquisition you can see the proportion of missed samples in the Trial list as one of the System Variables.
tip When acquiring from video files, in the Playback Control window select the option DDS (Detection Determines Speed). This way, no samples will be missed. See DDS (Detection determines speed)
Factors affecting the occurrence of missing samples
The following factors may cause the processor load to be too high:
▪Computer memory, processor speed and video graphics card capacity: See the system requirements. In general, using a computer with a dual-core CPU helps you to work with higher sample rates than normal computers do.
▪Other programs installed: Do not install other video software (for example, video editing programs, DVD burning software), because this can interfere with EthoVision XT's video processing and cause a reduction in performance.
▪Other programs are running: Make sure you shut down all other programs, including those running in the background such as e-mail programs and virus scanners. These are usually shown in the System Tray in the bottom-right corner of your screen.
▪Image resolution: For live video tracking, In the Experiment Settings you can choose the resolution for your video image.
▪Size of arenas: Make arenas as small as possible (but including the entire area the animal can be in).
▪Number of arenas: If you track live and use more than four arenas in a trial, check first that no samples are missed. If the number of missed samples is too high, first make a video file, then track from that. More generally, if you track from video files the number of arenas is never a problem as long as you select DDS (Detection determines speed).
When making detection settings, you could start with making an arena definition with only one area which speeds up the detection process. After you have finished configuring detection settings for one arena, you can add the others to the arena definition.
▪Display options: You can decrease processor load by minimizing the number of Track Features to be displayed (see View the detection features on the video window).
▪Detection method: If possible, use the Gray scaling method which requires less processor load than Static subtraction. Static subtraction requires less processor load than Dynamic subtraction and Differencing. See Advanced detection settings: Method