Real-time tracking of fast-moving object in occlusion scene

Tracking the fast-moving object in occlusion situations is an important research topic in computer vision. Despite numerous notable contributions have been made in this field,few of them simultaneously incorporate both object's extrinsic features and intrinsic motion patterns into their methodologies,thereby restricting the potential for tracking accuracy improvement. In this paper, on the basis of efficient convolution operators(ECO) model, a speed-accuracy-balanced model is put forward. This model uses the simple correlation filter to track the object in real-time, and adopts the sophisticated deep-learning neural network to extract high-level features to train a more complex filter correcting the tracking mistakes, when the tracking state is judged to be poor. Furthermore, in the context of scenarios involving regular fast-moving, a motion model based on Kalman filter is designed which greatly promotes the tracking stability, because this motion model could predict the object's future location from its previous movement pattern. Additionally,instead of periodically updating our tracking model and training samples, a constrained condition for updating is proposed,which effectively mitigates contamination to the tracker from the background and undesirable samples avoiding model degradation when occlusion happens. From comprehensive experiments, our tracking model obtains better performance than ECO on object tracking benchmark 2015(OTB100), and improves the area under curve(AUC) by about 8% and 32% compared with ECO, in the scenarios of fast-moving and occlusion on our own collected dataset.

Physical mechanism of mind changes and tradeoffs among speed,accuracy, and energy cost in brain decision making:Landscape, flux,and path perspectives

Cognitive behaviors are determined by underlying neural networks. Many brain functions, such as learning and memory, have been successfully described by attractor dynamics. For decision making in the brain, a quantitative description of global attractor landscapes has not yet been completely given. Here, we developed a theoretical framework to quantify the landscape associated with the steady state probability distributions and associated steady state curl flux, measuring the degree of non-equilibrium through the degree of detailed balance breaking for decision making. We quantified the decision-making processes with optimal paths from the undecided attractor states to the decided attractor states, which are identified as basins of attractions, on the landscape. Both landscape and flux determine the kinetic paths and speed. The kinetics and global stability of decision making are explored by quantifying the landscape topography through the barrier heights and the mean first passage time. Our theoretical predictions are in agreement with experimental observations： more errors occur under time pressure. We quantitatively explored two mechanisms of the speed-accuracy tradeoff with speed emphasis and further uncovered the tradeoffs among speed, accuracy, and energy cost. Our results imply that there is an optimal balance among speed, accuracy, and the energy cost in decision making. We uncovered the possible mechanisms of changes of mind and how mind changes improve performance in decision processes. Our landscape approach can help facilitate an understanding of the underlying physical mechanisms of cognitive processes and identify the key factors in the corresponding neural networks.

Exploration of unpredictable environments by networked groups

Information sharing is a critical task for group-living animals. The pattern of sharing can be modeled as a network whose structure can affect the decision-making performance of individual members as well as that of the group as a whole. A fully connected network, in which each member can directly transfer information to all other members, ensures rapid sharing of important information, such as a promising foraging location. However, it can also impose costs by amplifying the spread of inaccur- ate information （if, for example the foraging location is actually not profitable）. Thus, an optimal net- work structure should balance effective sharing of current knowledge with opportunities to discover new information. We used a computer simulation to measure how well groups characterized by dif- ferent network structures （fully connected, small world, lattice, and random） find and exploit resource peaks in a variable environment. We found that a fully connected network outperformed other struc- tures when resource quality was predictable. When resource quality showed random variation, however, the small world network was better than the fully connected one at avoiding extremely poor outcomes. These results suggest that animal groups may benefit by adjusting their informa- tion-sharing network structures depending on the noisiness of their environment.