Utilizing Temporal Information in Deep Convolutional Network for Efficient Soccer Ball Detection and Tracking

Abstract

Soccer ball detection is identified as one of the critical challenges in the RoboCup competition. It requires an efficient vision system capable of handling the task of detection with high precision and recall and providing robust and low inference time. In this work, we present a novel convolutional neural network (CNN) approach to detect the soccer ball in an image sequence. In contrast to the existing methods where only the current frame or an image is used for the detection, we make use of the history of frames. Using history allows to efficiently track the ball in situations where the ball disappears or gets partially occluded in some of the frames. Our approach exploits spatio-temporal correlation and detects the ball based on the trajectory of its movements. We present our results with three convolutional methods, namely temporal convolutional networks (TCN), ConvLSTM, and ConvGRU. We first solve the detection task for an image using fully convolutional encoder-decoder architecture, and later, we use it as an input to our temporal models and jointly learn the detection task in sequences of images. We evaluate all our experiments on a novel dataset prepared as a part of this work. Furthermore, we present empirical results to support the effectiveness of using the history of the ball in challenging scenarios.

Object detection on robosoccer environment using convolution neural network

“…The work done by Menashe et al [3] detects a ball, without knowing the ball's position for which series of heuristic region of interest identification techniques and super-vised machine learning methods was used. Kukleva et al [10] proposes a CNN approach for detecting and tracking the robotsoccer ball, in their approach they use the history of frames instead of using the current frame to detect the ball. Loncomilla and Solar [11] proposed YoloSPoc which uses maximal activation convolutions descriptors in which good quality object proposals are done by YoloV3.…”

Scientific and Technological Challenges in RoboCup

“…Ball tracking is also a necessary part of the game. Utilizing the temporal information in the CNN, Kukleva et al (71) presented a system that uses spatiotemporal correlation to efficiently detect and track a soccer ball based on its trajectory. Felbinger et al (72) designed a CNN for ball detection using a genetic approach that optimized network hyperparameters, providing a costeffective inference on the NAO with a limited amount of training data.…”

Ball Tracking Based on Multiscale Feature Enhancement and Cooperative Trajectory Matching

“…Kamble et al [4,5] added a classification confidence branch to the detection model, obtained the detection bounding box, and divided the image blocks into three categories, balls, players, and backgrounds, making the positioning of balls more accurate in the tracking phase. Kukleva [6] used a method based on a fully convolutional neural network with automatic encoding and decoding structure: first, generate a ball candidate area on the original image [7], then calculate the distance between the candidate area and the truth ball, delete the wrong candidate area by setting a threshold, and detect the model output containing the confidence of the ball and the diameter of the ball. In the tracking stage, existing research mainly focuses on solving the problems of motion blur and deformation caused by the fast movement of balls and trajectory incoherence caused by occlusion and out-of-drawing.…”