Innovative Prediction of Pass Feasibility in Soccer via Player Orientation
Written on
In this research, a computational framework is presented, aimed at improving the performance of football teams by integrating orientation metrics into the analysis of passing events. The foundation of this study lies in Pep Guardiola's assertion that modern soccer necessitates that players first assess their positioning before controlling and delivering the ball. To maintain simplicity, the investigation is concentrated on passing events, which are pivotal moments where orientation is essential.
The proposed model amalgamates three distinct feasibility metrics to evaluate the probability of a successful pass to possible recipients. Player orientation data is derived using advanced Computer Vision techniques that analyze upper-torso poses in a 2D field. This innovative feasibility metric evaluates the alignment between the passer's and receiver's orientations. Another component assesses the proximity of defenders relative to the passing path, gauging the difficulty of delivering the ball to a particular player. Furthermore, the model considers the distances between offensive players to determine the likelihood of receiving the ball, as players in closer proximity have a better chance of success.
Results indicate that the integration of all feasibility measures yields superior performance compared to any individual measure, with orientation feasibility proving to be particularly beneficial. The model's efficacy is also enhanced by incorporating orientation data into existing advanced models for predicting pass outcomes.
Subsequent sections delve into a review of related studies and provide a comprehensive description of the computational model, including technical details. The findings, discussions, and potential combinations are explored, leading to optimistic conclusions regarding the incorporation of orientation metrics to enhance football analytics and improve team performance.
Proposed Pass-Orientation Model
The proposed Pass-Orientation Model seeks to predict the most probable recipient of a pass based on prior information indicating that a player is about to pass. It introduces a feasibility score that factors in player orientation alongside the spatial configuration of both offensive and defensive teams at a specific moment.
The method works on color video frames taken at discrete time intervals. The visible players in each frame, along with their body orientations, are analyzed. The orientation data is acquired using the previously mentioned technique, simplifying the notation for player positions and orientations while omitting time dependencies for clarity.
Let P be the player possessing the ball at time t, serving as the passer, while {Ri, i = 1, …, I} and {Dk, k = 1, …, K} denote the positions of visible teammates and defenders, respectively. The first set includes potential recipients for the ball at time t + ?t, where ?t signifies the passing duration.
To evaluate the feasibility of a pass from player P to receiver Ri, three factors are considered: (a) the body orientation of all involved players, (b) defensive pressure from defenders Dk on both P and Ri, and (c) the relative positioning of Ri concerning P. The feasibility measure F(i) for the pass event Hi (passing to receiver Ri) is computed by combining three individual scores: the orientation score Fo(i), the defenders score Fd(i), and the proximity score Fp(i).
The orientation score Fo(i) assesses how well-aligned the passer P and receiver Ri are, while the defenders score Fd(i) gauges the defensive pressure on both players. The proximity score Fp(i) measures the distance between P and Ri.
The most feasible pass Hˆ is determined by maximizing the feasibility measure F(i) across all potential receivers at the moment the passer kicks the ball.
By integrating player orientation data and considering both offensive and defensive configurations, the proposed Pass-Orientation Model aims to enhance the analysis of passing events, yielding valuable insights into team performance and assisting coaches in formulating optimal strategies.
Orientation
This paper also examines the significance of player body orientation during passing events and proposes a computational model to evaluate pass feasibility based on these orientations.
To ascertain each player's orientation at the time of a pass, a window of ±Q frames surrounding the pass moment is analyzed, with the median value from this window representing the player's orientation. An orientation-based feasibility measure is introduced, reflecting the geometrical relationships between the passer and potential receivers.
In this model, all potential receivers Ri are adjusted to an equidistant position Z from the passer while preserving the original angles between the passer and each receiver. An isosceles triangle TP is defined based on the passer's orientation, indicating the area from which the passer can effectively deliver the ball. Similarly, a triangle TRi is defined for each receiver Ri, representing their field of view from which they can receive a pass. The orientation-based feasibility score is calculated as the weighted area of intersection between triangles TP and TRi.
The weights in the feasibility score formula consider the relative positions of points within the computed triangles, diminishing for further positions to avoid overvaluing intersections. Different triangle heights are utilized to prevent players positioned behind the passer from influencing the intersection significantly, even when the passer is not looking in their direction.
Defenders Position
This section discusses how the positioning of defenders influences the decision-making process regarding passing events in football. The behavior of defenders is crucial in determining pass feasibility, even when a player is well-positioned and oriented. The authors propose a computational model that evaluates the feasibility of passes based on defenders' locations concerning both the passer and potential receiver.
Two primary feasibility scores related to defenders are considered: Fd,P(Ri) and Fd,R(Ri). Fd,P(Ri) assesses the feasibility of passing towards the angle ?(P, Ri), which is the angle between the passer P and receiver Ri. This score is derived from the Euclidean distances of the J nearest defenders to the passer, with J being a predefined parameter. The distance weights are set to highlight defenders near the passing line, indicating a higher risk for that specific pass.
Conversely, Fd,R(Ri) reflects the feasibility of the receiver Ri receiving the ball from the passer. To ensure independence, the J nearest defenders from the passer are excluded, and the closest defenders to the receiver are identified. The overall defenders’ feasibility score, Fd(Ri), is then defined as the product of Fd,P(Ri) and Fd,R(Ri), representing the likelihood of a pass event to a specific player given the defensive spatial arrangement.
Overall, the defenders-based feasibility model introduces an essential element to the passing decision-making process by incorporating the positioning and proximity of defenders to both the passer and receiver. This thorough analysis enhances the accuracy of pass success assessments in dynamic football scenarios.
Pairwise Distances
The next section focuses on how the distances between players influence the feasibility of passing options. Player positioning on the 2D field significantly impacts their chances of receiving the ball, with those nearer to the passer being more likely candidates for receiving a pass.
To quantify this proximity-based feasibility, the authors propose a measure denoted as Fp(Ri). The feasibility score Fp(Ri) is inversely related to the Euclidean distance between the passer P and potential receiver Ri. An exponential function is utilized to calculate this score, where d(P, Ri) represents the distance between the passer and receiver in the 2D field.
The rationale behind this proximity-based measure is to prioritize players who are physically closer to the passer, making them more likely to be selected as candidates for receiving the ball. This measure complements the previously discussed orientation and defenders-based feasibility metrics, providing a holistic analysis of the factors influencing the most viable passing options on the field. Incorporating pairwise distances into the computational model further refines pass decision-making in dynamic football contexts.
Combination
The research concludes by introducing a method to integrate the three independent feasibility measures derived from orientation, defenders’ positioning, and pairwise distances. The objective is to achieve a comprehensive and robust assessment of potential passes in football.
The proposed approach combines the three feasibility scores, emphasizing that even if the other two values are high, a low score in any one of the three features (orientation, defenders, or distance) signifies a risky pass. Essentially, the feasibility of a pass is dictated by its weakest aspect, ensuring that the overall score reflects the most limiting factor in the passing situation.
By synthesizing the three feasibility measures, the model can evaluate passing options comprehensively, accounting for various factors such as player orientations, defenders’ positions, and pairwise distances. This integration leads to a more sophisticated and accurate evaluation of passing possibilities on the football field, empowering coaches and analysts to make informed decisions about the most effective passing strategies for optimal team performance.
Results
The authors present experimental findings and analyses regarding the performance of the proposed pass-orientation model. The dataset encompasses 11 complete games from F.C. Barcelona, comprising 6038 pass events, each labeled with a binary flag indicating success or failure. The primary goal is to explore the correlation between appropriate player orientation and successful receptions, thereby enhancing the likelihood of creating scoring opportunities.
To assess the influence of player orientation, the paper introduces a baseline pass model (Fpd) that only employs the outputs from Fp (proximity) and Fd (defenders’ position) feasibility measures. The Top-X metric is utilized as the primary accuracy assessment tool, measuring the percentage of instances where the actual receiver (the intended recipient of the pass) ranks within the top X candidates predicted by the feasibility models. The analysis examines both Top-1 and Top-3 accuracy metrics under varying conditions, accompanied by histograms that visualize the distribution of candidate receivers’ rankings based on feasibility values.
The histograms provide insights into the model’s performance by juxtaposing successful (blue) and unsuccessful (orange) pass events. Each histogram bin illustrates the number of times the actual receiver is positioned as the nth best candidate receiver, with n ranging from 1 to 10 (excluding the goalkeeper). The bin height reflects the frequency of the actual receiver being ranked at a specific position in the candidate list suggested by the feasibility scores.
Orientation Relevance in Pass Feasibility
The researchers evaluate the importance of player orientation in the proposed pass feasibility model (F) by contrasting it with a baseline model (Fpd) that omits orientation. The findings reveal that factoring in orientation is pivotal in determining pass outcomes.
Table 1 displays the Top-1 and Top-3 accuracy metrics for both F and Fpd. The results clearly illustrate that the features integrated into the feasibility computation correlate strongly with passing success. The disparity in accuracy between successful and unsuccessful passes is notably greater when orientation is factored in, underscoring its critical role in passing feasibility. F outperforms Fpd in both Top-1 and Top-3 accuracy metrics by margins of 0.07 and 0.05, respectively, reinforcing the significance of orientation in pass feasibility assessments.
The paper also scrutinizes the individual performances of the three independent feasibility measures: Fp (proximity), Fd (defenders’ position), and Fo (orientation). Table 2 and Figure 5 present the findings, revealing notable insights. For successful passes, the histograms for all three components exhibit similar distributions, indicating their collective significance. Fp assigns higher values to the top bins, suggesting that passing to players far away from the ball is less likely. Conversely, Fd and Fp are more relevant for unsuccessful passes, with Fd being particularly influential. This suggests that passing to a well-defended player is more likely to result in a turnover. Fo resembles Fp in distribution but is more evenly spread.
The paper highlights the importance of consolidating all three feasibility measures (F) for comprehensive contextualization. This combination maintains the high Top-1 and Top-3 accuracy metrics of Fp while preserving the impact of Fd in distinguishing between successful and unsuccessful passes. The research successfully creates a feasibility measure (F) that comprehensively integrates player orientation, distance, and defenders’ positioning, resulting in improved pass prediction and decision-making in football gameplay. The analysis outcomes validate the effectiveness of the proposed pass-orientation model in enhancing team performance and strategic planning during matches.
Players’ Field Position / Game Phase
This section examines how orientation as a feasibility measure influences different player positions and game phases in football. Players are categorized as defenders, midfielders, and forwards, with the feasibility measure applied to assess their pass success rates. Additionally, the orientation's effect is analyzed based on the passer's position relative to the defensive team's arrangement, leading to the identification of three offensive play phases: build-up, progression, and finalization.
The results presented in Figure 6 and Table 3 indicate that midfielders are the most influenced by orientation in terms of the feasibility measure. When orientation is included, both Top-1 and Top-3 accuracy metrics improve by 0.10, while maintaining a consistent difference in success rates between successful and unsuccessful passes for midfielders. Defenders, however, are less affected by orientation, typically making secure passes within their defensive zones, where orientation is less critical. Forwards experience some influence from orientation, but they engage in fewer passing interactions, leading to a higher turnover risk and potential rewards in their area.
The study further explores game phases based on the passer's position in relation to the defensive team's configuration. Three phases—build-up, progression, and finalization—are identified and represented in Figure 7.
The analysis shows that the impact of orientation is particularly pronounced in the progression phase, with substantial improvements in both Top-1 and Top-3 accuracy metrics, exceeding 0.2. In the build-up and finalization phases, which represent lower and higher risk scenarios, respectively, including orientation also enhances pass accuracy metrics.
Overall, the research underscores the significance of player orientation in determining pass feasibility across various player positions and game phases. By incorporating orientation as a key feature in the proposed pass feasibility model, the study yields valuable insights for optimizing passing decisions, improving team performance, and strategically planning offensive plays in football.
Combination with Expected Possession Value
The authors investigate the integration of player orientation within the existing EPV model, which aims to assign value to individual football actions, especially regarding pass probability modeling. The current EPV model does not consider player body orientation, resulting in somewhat accurate yet improvable outcomes. An illustration in Figure 9 highlights this limitation, where the EPV model fails to penalize a risky pass due to the passer’s limited field of view, leading to potential inefficiencies.
To address this shortcoming, the paper proposes merging the orientation-based feasibility measure Fo with the output maps from the original pass probability model (VP) or the EPV model (VE). However, a challenge arises due to the dimensional misalignment between the output maps (discretized field positions) and the individual feasibility values for potential receivers (10 candidates). To resolve this, a geometrical approach is employed, integrating the probability/EPV values over relevant areas extending from the passer to each candidate receiver.
The final individual value for receiver Ri, denoted V(Ri), is derived by integrating the probability/EPV values over a disc Qi of radius q and a tubular region Si of fixed width s, both centered at the 2D field position of the receiver. These regions Qi and Si facilitate capturing the significance of the pass trajectory. This method can be applied to both types of maps (VP or VE), enhancing the individual probabilities/expected values by incorporating player orientation Fo.
The results in Table 5 and Figure 11 demonstrate that incorporating orientation significantly boosts accuracy across all scenarios, particularly achieving a nearly 0.1 increase in top-1 accuracy for the pass probability model. Orientation also enhances the EPV model's raw performance, raising accuracy by 0.07 in top-1 accuracy, especially for cases where players are outside the passer’s field of view.
The research concludes that incorporating orientation into the current EPV framework can lead to a more precise model, providing better insights into the decision-making process and optimizing individual/team performance in football. The integration of orientation-based feasibility with EPV’s pass probability modeling illustrates the potential for enhancing existing models to deliver more nuanced and context-aware analyses of passing strategies and decisions on the field.
Conclusions
This research paper introduces an innovative computational model designed to assess the feasibility of passes in football matches. A key contribution of this model is the integration of orientation data, which is directly extracted from video frames using pose estimation techniques. The study finds that orientation data is vital in players' decision-making processes and is closely linked to play outcomes.
The orientation feasibility is calculated using a geometrical approach that considers the positions and orientations of offensive players on the field. This data is then combined with two additional estimations: one based on the defenders' locations concerning potential receivers and another based on the distances between players.
Moreover, the researchers examine how their model's output can be integrated with existing pass probability and Expected Possession Value (EPV) models. The results indicate promising advancements, suggesting that state-of-the-art methodologies can be enhanced by incorporating orientation data.
Future work will involve exploring the applicability of this model in other sports, potentially extending the passing feasibility analysis to the entire field. Additionally, the authors propose leveraging orientation as a core feature for recognizing team actions, which could optimize tactical strategies based on spatial offensive configurations. This research paves the way for improving decision-making and strategic planning in football and potentially other team sports through the use of player orientation data.