cows are dictated by their health condition and their physical environment, behaviours such as their frequencies of drinking, eating, lying, and rumination. Regular observation of dairy cows' behaviours ensures their good health and early detection of an abnormal posture or activity can help control the ailment. Dairy cows that are physically and emotionally healthy produce more high-quality milk [1].At present, it needs a human to observe and analyse dairy cows' behaviours. This person has to be well trained and experienced. However, a human cannot manually observe every cow in a farm all the time. It is too laborious and costly. Consequently, several researchers have tried to develop an automated system that can recognize cows' behaviour with little human effort such as these studies [2][3][4][5][6].Two main approaches for tackling this problem are sensor-based technique and vision-based technique. Between the two approaches, vision-based technique is more suitable for detecting dairy cows' behaviours because it is less costly compared to the sensor-based approach. Sensorbased approach requires at least one sensor for every cow and a team of technical staff to operate the system. Moreover, there is a risk of reduced milk production because the cows may be annoyed by the sensor. The ultimate goal of our studies is to create an automated vision-based system for detection and analysis of dairy cows' behaviours.As the first step in this endeavour, we explored the possibility of detecting dairy cows in a free-stall barn by the two following vision-based methods: (i) Feature point matching method: finding point correspondences between the reference and the target images using Speeded Up Robust Features (SURF) [7], a performance scale and a rotation-invariant interest point detector and descriptor.Abstract Behaviours of dairy cows reflect their health and emotions. Behavioural analysis by video surveillance is an accepted technique for helping cow-keepers to spot their cows' health problems. To perform a behavioural analysis, the presence and location of the cows need to be detected first. In this study, we used feature point matching method and foreground detection method to detect them. Two experiments were conducted in a dairy farm to detect cows in video frames recorded by a video camera installed over the top of a free-stall barn. A total of 800 frames of recorded cows' activities were captured. True and false positive and negative results were statistically confirmed by t test. We found that the accuracies of the feature point matching and foreground detection methods were 38.55 and 75.95 %, respectively; hence, for our setup, the foreground detection was a better method.

This paper presents a novel motion concatenation method for parametric motion synthesis techniques. First, motion groups are created based on the actions in each motion. We then extract all of the parameters that control the synthesized motions. To connect the motion groups, we propose a motion concatenation algorithm based on cubic Bczier interpolation that can be used to connect any pair of motions. All of the poses are pre-calculated before interpolation, so that the concatenated motions can be synthesized rapidly during the concatenation phase. Although there is no intersection region between the parameter spaces, the proposed method guarantees that transitions between motions can be generated for any consecutive motions, which is a problem found in existing methods.

Motion databases usually contain sequences of movements and searching these vast databases is not an easy task. Motion clustering can reduce this difficulty by grouping sample movements into various motion groups containing similar actions. The pose distance is often used as a feature during motion clustering tasks. However, the main weakness of this strategy is its computational complexity. Query motions are also required to cluster motion sequences. To address these problems, we propose a motion-clustering algorithm based on the use of kinetic energy to cluster sample motions. Our method does not require query motions during the clustering process, so the clustering results can be generated without supervision. Our experimental results confirmed that our proposed method delivered comparable per formance to pose distance-based methods, while its computational complexity was significantly lower than that of existing methods.