1. Introduction
Boxing is not only a popular combat sport with a long tradition, but it has recently become a popular fitness trend as well [
1], with everyday people participating in boxing-related fitness classes hoping to improve their aerobic capacity, reduce their body fat percentage, etc. As this type of training reduces obesity [
2], increases cardiovascular health [
2,
3], and improves aerobic capacity [
4], it is no surprise that people seek to participate in this type of high-intensity training. However, as with most types of training, the volume and intensity of training are two of the main factors to consider when designing and implementing a training program. In traditional exercises, such a strength training, it is easy to prescribe a set number of repetitions with a specific load. However, the non-structured, repetitive, and highly dynamic nature of punching makes it difficult to prescribe or quantify training loads. Therefore, it would be advantageous to use technology that could provide objective data to quantify training volume while punching.
Among the available technology today, accelerometers can be used to detect punch type and provide data regarding punch force, velocity, power, and other measures that can help quantify punching training structure, volume, and intensity [
5,
6,
7]. Other technologies available include high-frame-rate video capture [
8]. Although accelerometers have been tested during punching [
7,
9], they have primarily included custom-made devices and algorithms that likely are not used by commercial users. Furthermore, the data collected in those studies are unique to specific audiences (e.g., for scoring and judging strikes, wrist angles, and other variables that everyday users likely are not interested in). Additionally, accelerometers that are invented specifically for the purpose of collecting punching data provide post-workout summaries, known as punch trackers, can even provide instantaneous feedback [
9,
10], which has been shown to play a role in maximizing acute performance [
11] and increasing motivation [
12,
13,
14]. Although these devices are interesting, and the data they provide could be useful, there is a lack of published data to support their validity, likely due to the novelty of the devices.
Although not publicly available, it can be assumed that the algorithms of these punch trackers slightly differ between manufacturers [
15]. Furthermore, since the resultant data are based on accelerometry, it is possible that punches thrown with slightly different techniques or trajectories may not be recognized by the punch trackers, reducing their accuracy in terms of quantifying training volume or providing objective feedback. Along these lines, it is possible that the same punches thrown by untrained punchers with less technically correct movement may not be recognized as well as in trained punchers who have better and possibly more consistent punch techniques, especially for more complex movements that require greater coordination (e.g., hooks versus jabs) [
16]. Although consumers use these punch trackers during training, their validity has not been assessed in an independent laboratory, which could provide additional information in terms of their ability to function well in real-world settings. Therefore, the purpose of this study was to compare four commercially available punch trackers to determine how well they could recognize specific types of boxing punches thrown by trained punchers and untrained punchers during shadow boxing. This study hypothesized that (I) the punch trackers would better register the total number of punches thrown by trained punchers compared to untrained punchers; (II) simple punches (lead and rear straights) would be detected with higher accuracy than more complex punches, such as lead and rear hooks and uppercuts; and (III) punch recognition would decrease throughout a consecutive sequence due to the hands not “resetting” after each punch, which may not align with the punch algorithms within the devices.
3. Results
The total punch counts for each punch type and each punch tracker are shown in
Table 3. The punches in the video recordings were all able to be identified as a specific punch type by the researcher, indicating that the movement pattern of the subject’s hands matched what would be expected for such a punch type. Therefore, the researcher judged that all punches were performed within the expected movement patterns, but it was possible that subjects performed an LH instead of an LUC (for example). In these cases, the LH was the actual punch thrown, which was recognized by the punch tracker. The results of MPE, MAPE and TOST for CPT, EPT and HPT for all participants and each group (TR and UNTR) are shown in
Table 4. The linear mixed-effects model indicated that the percentage error was significantly affected by punch type (
p < 0.001) and training experience (
p = 0.007). Specifically, the post hoc analysis revealed that the percentage error was lower for straight punches (lead and rear) compared to hooks and uppercuts (
p < 0.001) for all three punch trackers (
Table 5).
The sensitivity and specificity for CPT and HPT for recognizing individual punches (LS, RS, LH, RH, LUC and RUC) are present in
Table 6. The logistic regression with mixed effects indicated that there was a significant negative effect of the order within a sequence (
p < 0.001 for CPT and
p < 0.001 for HPT) and positive effect of the position within a round (
p = 0.024 for CPT and
p = 0.003 for HPT). In other words, the earlier within a sequence and the later within a round the punch was thrown, the better it was recognized.
Using the straight lead as an example, sensitivity is the proportion of straight lead punches that were correctly recognized as such, and specificity is the proportion of non-straight lead punches that are recognized as non-straight lead punches (but not necessarily recognized correctly). Sensitivity and specificity values as close as possible to one are desired.
4. Discussion
The main findings are that (I) the CPT, EPT, and HPT all detected punches with more accuracy in TR than UNTR participants; (II) the CPT, EPT, and HPT were all better at detecting straight punches compared to uppercuts and hooks; and (III) the successful recognition of punches with CPT and HPT depended on the order of boxing punches, with earlier punches in a sequence being recognized better. The same may or may not have occurred with the EPT, but the device does not allow for data to be exported, meaning individual punch data, such as the order of individual punches, could not be analyzed.
Based on the data presented, which supported the first hypothesis, participants with combat sport experience can use CPT, EPT and HPT to detect the total number of punches per session with reasonable accuracy. However, in UNTR participants, the EPT and HPT underestimated the total punch count, meaning that the CPT may be a better choice for untrained punchers in this regard. Considering that the punch trackers used in this study likely have unique algorithms for identifying different punch types [
1,
15] the technical implementation of each punch likely played a major role in the ability of each punch tracker to correctly register every punch. Since the EPT and HPT underestimated the total punch count in UNTR, it is possible that the thresholds needed to register a punch were not met, which could be a result of greater variability in the punch technique in UNTR compared to TR [
16].
Considering the punch technique, the second hypothesis was also confirmed as the CPT, EPT, and HPT were all able to better detect straight punches than hooks and uppercuts (
Table 5). Specifically, the HPT had better sensitivity (recognition) than CPT for straight punches. However, the CPT was better than the HPT for correctly detecting hooks and uppercuts. Since hooks and uppercuts are delivered in a curved swinging motion with a vertical drop in the initiation of the punch, they are more technical and complex than straight punches [
16]. Therefore, the UNTR punchers likely were unable to maintain the proper technique, resulting in worse upper cut and hook detection by the punch trackers compared to TR. Considering the strict technical requirements of hooks and uppercuts compared to straight punches, the likelihood of a “false-positive” decreases for hooks and upper cuts, which is supported by a greater specificity for hooks and uppercuts than straight punches (
Table 6). In short, if a punch tracker registered a hook or uppercut, it likely actually was a hook or uppercut, since a straight punch would likely not include an arcing pattern, even for the most inexperienced punchers.
The third hypothesis was also confirmed because regardless of training experience, increasing the number of punches in a sequence negatively influenced the recognition of punch type as the order of punches progressed. Although it is possible that the participants were able to focus better on the first punch of a multipunch sequence, losing their focus as the sequence progressed, the more likely explanation is that the first punch was performed from a static position. For subsequent punches, the punch trackers may not have registered returning to the start position, which may reduce their ability to correctly detect the next punch. Furthermore, the technique of transitioning from one punch to the next simply may not have corresponded with the movements that were expected in the respective algorithms. Contrary to the negative effect of the order of punches within a sequence, as each round of shadow boxing progressed (i.e., after multiple sequences), the CPT and HPT better recognized punch types in both TR and UNTR participants. It is possible that there was a learning effect, which has previously been shown to increase punch force and velocity after only 15 min of practice [
17], but such rapid skill acquisition would have likely occurred only in UNTR. Nevertheless, the present data do not allow for such a conclusion, and the most logical explanation for the increased recognition over time is the pyramid nature of the protocol. Subjects performed a series of single punches, followed by two-punch combinations, three-punch combinations, two-punch combinations, and finished with single punches. As such, the latter punches of the round were in fact single punches, meaning that the number of punches per sequence likely plays a greater role in punch recognition than the overall time spent punching. Therefore, any possible learning effect may be negligible in such a short time period, and the transitions between punches (i.e., the lack of coming back to a static starting position) likely make it difficult for the punch trackers to correctly identify multiple punches in sequence.
In addition to the main findings above, there are many factors to consider when interpreting the data of the present study. First, the maximum number of punches in a sequence was three. Considering the negative effect of the number of punches in a sequence on proper recognition, the data from each punch tracker would likely differ, and possibly worsen, if the number of punches per sequence increased past three. Thus, future research should investigate the punch recognition ability of these trackers in situations where many punches are performed in sequence. Second, the EPT only provides average data from the whole session for each punch type (
Table 2), meaning that punch-by-punch analyses are not possible, which is a factor to consider depending on the user’s needs. Third, due to an insufficient amount of data (
Table 3), the SPT data were not analyzed. Therefore, it cannot be concluded that SPT is not reliable for detecting punch types, as the SPT used in the present study may have been defective. On the other hand, it may not have been defective, and future research should aim to determine how the SPT performs under different conditions. Fourth, the CPT and HPT likely provide the most valid data for detecting and recognizing punch types. For detecting the total punch count, the CPT and HPT are both acceptable, particularly the CPT for participants without any training experience, and the HPT for more experienced participants. Although CPT, HPT, and EPT were better at detecting straight punches than hooks and uppercuts, a punch-by-punch analysis showed that the CPT and HPT not only detected but successfully recognized straight punches better than hooks and uppercuts (the EPT does not allow for such an analysis).
The CPT and HPT can both be used to evaluate shadow boxing with multiple punches, but single punches would likely be recognized more accurately. Lastly, the protocols were performed under standardized conditions, with a specific count of punches and combinations, all while boxing without an opponent. Thus, altering any combination of these conditions may affect the ability of these punch trackers to provide valid punch data, and future research should investigate these effects.