In recent years, sports analytics is becoming more and more demanding in the football area. Different methods for sports analytics are emerging in this area. Convolutional Neural Network based visual object detectors and histogram-based detectors are two prominent methods used in this area now. In this paper, we focus on a particular sub-domain: player detection and tracking in football matches. We develop different systems based on histogram and CNN respectively and carry out experiments for players detection and tracking in real-time football match videos. Instead of using open-source datasets for training, such as Pascal VOC and ImageNet, we use the collected data from real-time football matches provided by PIERO team, which makes it more difficult due to strong motion blur and large variance of image quality. With the help of transfer learning, we propose a customized in-house CNN based system. Experiments are conducted to evaluate our customized in-house CNN system against the sequence adaptive histogram-based system and other benchmarks. Results show that our in-house CNN system outperforms the others in terms of mean Intersection Over Union, precision rate and Mean Average Precision. Furthermore, we combine the CNN based system with the histogram-based sequence adaptive system as a post-processor to take advantage of the object’s visual appearance characteristic. The combined system is evaluated against our pure in-house CNN based system and CNN-Simple Online and Realtime Tracking (SORT) system, and it achieves better accuracy in terms of F1 and ITP scores.