MambaKick: Early Penalty Direction Prediction from HAR Embeddings

TL;DR

MambaKick is a learning-based framework that predicts penalty shot directions in soccer using pretrained HAR embeddings and lightweight temporal models, enabling low-latency decision support.

Contribution

It introduces a novel combination of pretrained human action recognition embeddings with efficient temporal modeling for sports intention prediction.

Findings

Achieves up to 53.1% accuracy for three-class prediction

Improves or matches baseline performance across HAR backbones

Demonstrates practical low-latency prediction in real-world sports footage

Abstract

Penalty kicks in soccer are decided under extreme time constraints, where goalkeepers benefit from anticipating shot direction from the kickers motion before or around ball contact. In this paper, MambaKick is presented as a learning-based framework for penalty direction prediction that leverages pretrained human action recognition (HAR) embeddings extracted from contact-centered short video segments and combines them with a lightweight temporal predictor. Rather than relying on explicit kinematic reconstruction or handcrafted biomechanical features, the approach reuses transferable spatiotemporal representations and utilizes selective state-spare models (Mamba) for efficient sequence aggregation. Simple contextual metadata (e.g., field side and footedness) are also considered as complementary cues that may reduce ambiguity in real-world footage. Across a range of HAR backbones, MambaKick consistently improves or matches strong embedding baselines, achieving up to 53.1% accuracy for three classes and 64.5% for two classes under the proposed methodology. Overall, the results indicate that combining pretrained HAR representations with efficient state-space temporal modeling is a practical direction for low-latency intention prediction in real-world sports video. The code will be available at GitHub: https://github.com/hvelesaca/MambaKick/