Table 3.

Summary of the sensors selected, its specifications (size and weight), the body’s placement, variables analyzed, and the main findings.

Source	Sensor	Specifications	Body’s Placement	Variables	Findings
Dadashi et al. [21]	Three wireless waterproofed IMUs (each with a 3D accelerometer and a 3D gyroscope).	n/a	Both wrists and lower back.	Duration of the arm pull, duration of the arm-push, duration of the recovery, and index of coordination.	It was possible to validate the estimation of front-crawl temporal phases extracted from IMUs. The automatic phase detection method provides timely feedbacks that can be used by sport scientists and coaches. This approach can be modified to fit event detection problems in other types of locomotion.
Dadashi et al. [22])	One waterproofed IMU (with a 3D accelerometer and a 3D gyroscope, a battery, a memory unit, and an MCU).	Dimensions: 50 × 40 × 16 mm. Weight: 36 g.	Lower back.	Velocity.	This study reported an accurate estimation of velocity that relies on the learning of the mapping between compact representation of the inertial signals and target cycle velocity measured by a tethered reference. The method provides immediate feedback on the variability of swimmer’s performance that a coach can use to provide tailored feedback to the swimmer during training.
Engel et al. [23]	One IMU (with a 3D accelerometer and a 3D gyroscope).	n/a	Lower back.	Phases of the arm stroke, roll angle, and angular velocity of the hip.	It was shown that athletes with different skill levels show the same characteristics in their IMU data, which is fundamental for the development of algorithms and for the analysis of the front-crawl swimming stroke, not only considering frequency and number of strokes, but also access to intra-cyclic parameters.
Fantozzi et al. [24]	Five triaxial IMUs equipped with an accelerometer and gyroscope.	n/a	Head, forearms, and shanks.	Wrist entry, head entry, head exit, and leg downbeat.	A protocol for integrated analysis of stroking, kicking, and breathing using inertial sensors in front-crawl swimming was developed and validated in comparison with a video-analysis technique. All investigated accuracy parameters highlighted strong agreement with the gold standard.
Hagem et al. [25]	Acceleration sensors with an MCU and memory for data recording. A system with pre-programmed feedback (audio, tactile, and visual).	n/a	Wrist and head (eyes—goggles).	Phases of the arm stroke, and stroke rate.	The wearable data acquisition, processing, and feedback system was designed, implemented, and tested based on visible light communication in order to give a real-time feedback to a swimmer during swimming. Acceleration data were used for stroke rate determination and optimum transmission time in the stroke cycle.
Hagem et al. [26]	An accelerometer sensor and an MCU at the receiver side that saves the data, decides based on preset conditions, and sends feedback to a display mounted on the goggles.	n/a	Wrist and head (eyes—goggles).	Stroke count, time, stroke rate, stroke length, velocity, and stroke duration.	A short-range optical wireless transceiver was designed and implemented for real-time swimmer feedback applications. The system was based on using an encoder and a decoder with an optical transceiver. The information transmitted was the time duration of one complete stroke, which was updated every stroke and presented to the swimmer using an RGB LED mounted on the goggles.
Hagem et al. [27]	Transmitter that includes an MCU unit with memory. Three-axis accelerometer, power supply, and battery charging circuit are included in the circuit board.	Dimensions: transmitter—35 × 35 mm2; receiver—45 × 30 mm2.	Wrist and head (eyes—goggles).	Stroke rate.	A second-generation system was designed and implemented. The system was tested with different swim speeds (slow and fast) and different strokes (freestyle, backstroke, breaststroke, and butterfly) to validate the system. These experiments were used to optimize the system and verify that the complete system was viable under different conditions, strokes, and swimmers.
Jeng [28]	Charging and power supply circuits, as well as MCU and IMU data storage circuits.	Dimensions: 53 × 29 × 5 mm. Weight: 15 g (before water case).	Head.	Pitch and roll angles of the butterfly stroke breathing pattern.	To investigate the influence of breathing motions on swimming speed during the butterfly stroke, an IMU was developed. It was showed that significant interaction effects exist between age and average breathing time, significantly influencing swimming efficiency. It also indicated that significant interaction effects exist between gender and the number of breaths taken and between gender and average maximum breathing angle. These results demonstrate the efficacy of the proposed IMU, which could be effectively applied to help coaches and researchers analyze and enhance swimmers’ performance.
Le Sage et al. [29]	An MCU in combination with an RF transceiver, and a tri-axis accelerometer and dual-axis gyroscope.	The packaging (containing all systems) has a combined mass of 110 g. Dimensions: 15 × 9 cm.	Lower back.	Stroke count, stroke rate, and stroke duration.	A novel approach to monitoring free swimming performance with embedded real-time filtering and signal processing was developed. The system exhibits many advantages over current analysis techniques since it offers the opportunity to provide feedback to coaches, performance analysts, and sports scientists in real time, as well as more rapid feedback to swimmers.
Lee et al. [30]	Apple Watch S2 and Garmin Fenix 3HR.	n/a	Wrist.	Lap count, stroke count, energy expenditure.	The error rate of lap counting and stroke counts at various swimming speed were within 10% for Apple and about 20% for Garmin. The criterion measurements and a 95% equivalence test showed that the lap counts and the strokes counts recorded by Apple were within the equivalence zone for all of the exercise intensities measured. Bland–Altman plots showed confidence intervals with relatively small deviations in lap counts and the stroke counts for Apple, and energy expenditure for Garmin. However, the error rate of estimating energy expenditure was higher for Apple than for Garmin. Apple and Garmin wearable watches accurately measure lap counts and stroke counts. However, the accuracy of estimating EE is poor at slow to medium swimming speeds.
Mangia et al. [31]	Seven IMUs.	Dimensions: 48.4 × 36.5 × 13.4 mm. Weight: <22 g.	Thorax, arms, forearms, and hands.	Phases of the arm stroke, arm pull duration, arm push duration, non-propulsive phase duration, and stroke rate.	The use of IMUs can provide several advantages over more expensive and bulky systems, such as (1) simpler and faster setup preparation; (2) less time-consuming processing phase, and (3) the chance to record and analyze a higher number of strokes without limitations imposed by the camera’s volume of acquisition.
Pan et al. [32]	A module that gathers linear acceleration values from the accelerometer every n millisecond.	n/a	Palm of the hand.	Stroke count and stroke identification.	A swimming analysis scheme to count and identify swim strokes using an accelerometer was proposed. The stroke analysis phase identifies stroke styles and counts strokes by finding correlated segments, which can be taken as swimmers’ strokes. It was implemented the designed scheme on a waterproof Android platform. The experiment results indicate that the designed scheme can effectively identify stroke styles and count strokes with more than 87% and 94% accuracies on average, respectively.
Pla et al. [33]	TritonWear (triaxial accelerometer, triaxial gyroscope, and triaxial magnetometer).	n/a	Head.	Lap time, stroke count, velocity, stoke rate, stroke length, and stroke index.	The accuracy of spatial–temporal variables with the use of the TritonWear was high in international open-water swimmers. This device may help coaches to analyze spatial–temporal variables during swim training to determine their relationship with performance. The ease of use, the good accessibility, and the ease in understanding the results of this device allow the coaches to give quick feedbacks and advice to the swimmer during swim training.
Rowlands et al. [34]	Gyroscope.	n/a	Lower back.	Angular velocity of the body roll.	The body roll velocity was captured from the gyroscopic sensor and was used to visualize the time-series data. The visualization techniques that were investigated were time series overlay, phase space portraits (two different methods), ribbon plots, and wavelet scalograms. Obvious differences were observable in all the visualization methods. It was found that all the methods were able to give useful information on the consistency of the stroke cycle. Each of the visualization techniques also showed that the consistency was higher in the elite swimmer than the sub-elite swimmer which was expected. Therefore, these techniques do show merit due to the extra information that can be provided on the swimming action.
Shell et al. [35]	TritonWear (triaxial accelerometer, triaxial gyroscope, and triaxial magnetometer).	n/a	Head.	Distance, stroke count, velocity.	The wearable device investigated in this study does not accurately measure distance, stroke count, and velocity swimming metrics, or detect stroke type. Its use as a training monitoring tool in swimming is limited.
Slawson et al. [36]	Accelerometer.	n/a	Lower back.	Acceleration on the turn approach, rotation, and glide.	Using vision data, it was possible to determine turning phases based on acceleration characteristics. This enabled more complete analysis of turning performance as the approach, rotation, and glide could be individually identified. This is a proactive method to alert users to events, rather than the coach or biomechanist, who have to analyze every output to judge whether something is of significance.
Slawson et al. [37]	MCU, analogue-to-digital converter with an associated sensor, triaxial accelerometer, biaxial gyroscope, digital interface (to enable the connection of additional memory), crystal oscillator, radio components, and power.	Dimensions: 90 × 40 mm.	Lower back.	Acceleration on the turn approach, rotation, and glide.	Using the visual data, it was possible to understand the turning phases in acceleration space. Rotation information, last stroke to wall, and first stroke after the turn timing could be automatically distinguished from the captured acceleration data. Turning information, automatically cropped from the raw data stream within these stroke time limits, enabled a more complete analysis of turning performance than what has been previously possible, as the approach, rotation, and glide features could be individually identified and quantified.
Stamm et al. [38]	Triaxial accelerometer, triaxial gyroscope, and radio frequency capabilities.	n/a	Lower back.	Stroke rate, stroke duration, and velocity.	Considering the small and light weight of the used sensor, it can be nearly used during every swimming session. This offers the opportunity to all athletes and coaches to record as many swimming sessions as they want without complex and bulky equipment. It allows arm symmetry investigations in different ways (stroke rate, acceleration, and velocity) and offers the possibility of keeping track of training progress or injury recovery. Furthermore, it brings up the opportunity to identify symmetry problems in swimming styles and helps to adjust the swimmers´ style if necessary.

3D—three dimensions; IMU—inertial measurement unit; LED—light-emitting diode; MCU—microcontroller unit; RF—radio frequency; RGB—red, green, blue; n/a—not applicable or not disclosed.