Real-time compressive tracking is proposed in this paper, which uses a data-independent appearance model based on features extracted from a multi-scale image feature space using non-adaptive random projections. The algorithm compresses samples of foreground targets and background using a sparse measurement matrix, and formulates the tracking task as a binary classification problem using a naive Bayes classifier with online updates in the compressed domain. The algorithm is efficient and runs in real-time, performing well against state-of-the-art algorithms in terms of efficiency, accuracy, and robustness. The paper discusses the challenges of object tracking, including pose variation, illumination change, occlusion, and motion blur. Existing online tracking algorithms often update models with samples from recent frames, but face issues such as data dependency and drift. The proposed algorithm addresses these issues by using a sparse measurement matrix that preserves the structure of the image feature space and allows efficient feature extraction. The algorithm is generative as it can represent the object well in the compressed domain and discriminative as it uses these features to separate the target from the background. The algorithm uses a sparse random projection matrix that satisfies the restricted isometry property, enabling efficient projection from the image feature space to a low-dimensional compressed subspace. Positive and negative samples are projected using the same sparse matrix and classified by a naive Bayes classifier. The algorithm is tested on 20 challenging sequences, achieving the best or second best results in terms of success rate and center location error. It is also robust to occlusion, pose variation, and scale changes, and performs well in sequences with significant pose changes and abrupt motion. The algorithm is implemented in MATLAB and runs at 35 frames per second on a standard computer. The results show that the proposed algorithm is efficient, accurate, and robust, making it a promising approach for real-time object tracking.Real-time compressive tracking is proposed in this paper, which uses a data-independent appearance model based on features extracted from a multi-scale image feature space using non-adaptive random projections. The algorithm compresses samples of foreground targets and background using a sparse measurement matrix, and formulates the tracking task as a binary classification problem using a naive Bayes classifier with online updates in the compressed domain. The algorithm is efficient and runs in real-time, performing well against state-of-the-art algorithms in terms of efficiency, accuracy, and robustness. The paper discusses the challenges of object tracking, including pose variation, illumination change, occlusion, and motion blur. Existing online tracking algorithms often update models with samples from recent frames, but face issues such as data dependency and drift. The proposed algorithm addresses these issues by using a sparse measurement matrix that preserves the structure of the image feature space and allows efficient feature extraction. The algorithm is generative as it can represent the object well in the compressed domain and discriminative as it uses these features to separate the target from the background. The algorithm uses a sparse random projection matrix that satisfies the restricted isometry property, enabling efficient projection from the image feature space to a low-dimensional compressed subspace. Positive and negative samples are projected using the same sparse matrix and classified by a naive Bayes classifier. The algorithm is tested on 20 challenging sequences, achieving the best or second best results in terms of success rate and center location error. It is also robust to occlusion, pose variation, and scale changes, and performs well in sequences with significant pose changes and abrupt motion. The algorithm is implemented in MATLAB and runs at 35 frames per second on a standard computer. The results show that the proposed algorithm is efficient, accurate, and robust, making it a promising approach for real-time object tracking.