Signal and Image Processing in Target Detection and Tracking
Signal processing is a field that mainly focuses on analyzing and synthesizing the signals. The signals can be sound or image. Signal processing techniques are used to improve transmission, subjective quality. These techniques can be applicable in many fields such as, video processing, image processing, control systems, etc. Image processing is the sub field of the signal processing . It is very vast field. So, by using these two fields we can do target detection and tracking.
Signal processing is a field that mainly focuses on analyzing and synthesizing the signals. The signals can be sound or image. Signal processing techniques are used to improve transmission, subjective quality. These techniques can be applicable in many fields such as, video processing, image processing, control systems, etc. Image processing is the sub field of the signal processing . It is very vast field. So, by using these two fields we can do target detection and tracking.
Do you know Target Detection?
Nowadays, computer vision is one of the hotspots of the artificial intelligence. while using the computer vision in our day-to-day life we must know about the basic topics. These topics contains image classification, semantic and instance segmentation and the easy but important topic i.e. object/target detection
Target detection is a computer technology, which is related with the image processing, computer vision and also with the signal processing. Using this we can identify any particular object or item in an image or in video. So, in simple words we can say that its an ability of a computer or that algorithm to detect the object from images.
In the initial stages this detection is done with signal processing. Then with modification many methods are introduced. These methods are very reliable and they give results with more accuracy than previous one within less time.
Techniques of doing Target detection
From past two decades, tremendous modifications have done in the target detection techniques. This field is very vast, so many researchers across the world gave their opinion about how we can detect any particular item/ target from the digital image or from the video. Total 23 papers were accepted as a good and applicable for target detection. These papers include many techniques such as Joint Transform Correlator (JTC) based recognition, synthetic aperture radar based recognition, wavelet based recognition, etc.
One of these papers, they are recognizing object through the Maximum Average Correlation Height filter also known as ‘MACH’ filter. But this filter, confuses the training model and leads to the poor discrimination of desired target from the clutter.
At the end they come up with the solution that they introduced two new metrices. The two metrices are All Image Correlation and Modified Image Similarity Measure, also known as ‘AICH’ and ‘MISM’. Now this advanced filter has better ability to execute clutter rejection. There are many such kind of advanced techniques to detect object in less time with accuracy.
How Target Tracking works?
We discussed the object detection earlier, now the next is to track that object. So, the object tracking is the simple process of just giving a unique id to the object and keeping the track of that object. By giving the unique id to that object it makes possible to us to count objects in an image or video , we can count how many times the particular object repeatedly comes in picture. The object tracking is the best while building any counter like, we in criminal investigation, the officers can count that how many times the criminal comes to a particular place. It helps to many other applications also.
The tracking methods are classified in three types such as kernel, point and silhouette based tracking. The kernel method is most used method among these types due its high accuracy. The point tracking method has less computational cost but when it comes to accuracy this method is not much useful.
Algorithm for Target Tracking
As we know, if we have problem then many experts come up with their solution with different ways. Similarly, target tracking has many algorithms to track the objects with more n more accuracy. The Centroid Tracking Algorithm is the well-known algorithm for target tracking.
It is easy to implement and very effective while tracking. This algorithm uses Euclidean distance method ,it calculate the distance between existing object centroids and new object centroid. This algorithm contains 5 steps as follows,
I. Computing centroids by accepting the bounding box coordinates
This algorithm assumes that we are passing detected objects in a set of bounding box in (x,y) coordinates. Every bounding box have its own centroid and we have to compute them . At last of this step, we gave a unique name or id to each bounding box.
II. Computing Euclidean Distance
This step we are calculation Euclidean distance between existing and new object centroids. This step is very important because while assigning new id to the centroid, we must need to determine that we can associate the new centroid of new bounding box with the existing centroid of bounding box. To solve this, we compute the Euclidean distance.
III. Updating (x,y) coordinates
After successfully computing Euclidean distance the question arises that how do we use these distances to match the objects and associate them. The basic assumption of this algorithm is that an object will potentially move between subsequent frames. But the main condition is The distance between two centroids must be smaller than the other objects. So that we can associate centroids with minimum distances between subsequent frames and with the help of this we can successfully build the object tracker.
IV. Register the New objects
For associating centroids we need small distance but ever you wonder if the points alone in the frame?
To solve this problem, we just simply register the new object. It means we are adding new object in our tracked object list. Then we assign new id to the object and stores the bounding box coordinates for the new object. Then again, we have to do all steps from step 2.
V. Deregister Old objects
Any object tracking algorithm needs to be able to handle when an object is disappeared. To handle this situation, we can deregister the objects. It totally depends on where our object tracker is implemented.
The use of signal processing in object detection tracking is less useful compared to use of image processing. In radar system, we can detect any threats which comes in the range. Whenever the threat comes in range it interrupts the signal sent by the machine. Due to this interruption, the machine detects and send message to the instructor about the threat. It is the best and well-known example of the use of signal processing in object detection and tracking.
The GPS is the well-known use of image processing in target detection and tracking. Another example is let’s say, in any crime investigation the officers has only a video as a proof. So, by doing the target detection techniques they can identify many other important things. Also, they can keep the track of the detected things. It makes easier to solve the case using these two techniques.
However, we need to concentrate on improving the accuracy level towards recognizing with limited trained objects as reference. Also, test the effectiveness of this approach on images of other objects. With improvements we can extract important things from a complex video sequence and can track them using various algorithms. Also, we need to improve the enhancement in detection techniques like they can detect the object with certain illumination changes, shadows, etc.
Thanks for reading!!
