# Kalman Filter with OpenCV (C++), Assertion failed (a\_size.width == len) in gemm

**URL:** <https://forum.opencv.org/t/kalman-filter-with-opencv-c-assertion-failed-a-size-width-len-in-gemm/6767>\
**Category:** C++\
**Tags:** ros\
**Created:** [December 16, 2021, 1:03pm UTC](https://forum.opencv.org/t/kalman-filter-with-opencv-c-assertion-failed-a-size-width-len-in-gemm/6767 "2021-12-16T13:03:07Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![diogofd8](https://avatars.discourse-cdn.com/v4/letter/d/bc8723/32.png) [@diogofd8](https://forum.opencv.org/u/diogofd8)\
**Post date:** [December 16, 2021, 1:03pm UTC](https://forum.opencv.org/t/kalman-filter-with-opencv-c-assertion-failed-a-size-width-len-in-gemm/6767/1 "2021-12-16T13:03:07Z")

</div>

'm trying to create a simple Kalman Filter tracker using the openCV library. I based on tutorials for similiar goals in: [http://opencvexamples.blogspot.com/2014/01/kalman-filter-implementation-tracking.html](http://opencvexamples.blogspot.com/2014/01/kalman-filter-implementation-tracking.html) [Object Tracking: Simple Implementation of Kalman Filter in Python](https://machinelearningspace.com/object-tracking-python/)

However, I’m getting an issue with matrix sizes. I’m confident that I’m setting them right and I’ve been around this for hours without being able to fix it. Any suggestions?

```auto
    cv::KalmanFilter KF(STATE_SIZE, MEAS_SIZE, CONTR_SIZE, CV_32F);

    // Initial State Matrix
    // [p_x] 
    // [p_y]
    // [p_z]
    // [v_x]
    // [v_y]
    // [v_z] 
    KF.statePre = cv::Mat::zeros(cv::Size(1, STATE_SIZE), CV_32FC1);
    KF.statePre.at<float>(0) = odometry.x;
    KF.statePre.at<float>(1) = odometry.y;
    KF.statePre.at<float>(2) = odometry.z;

    cv::setIdentity(KF.measurementMatrix);
    cv::setIdentity(KF.processNoiseCov, cv::Scalar::all(1e-4));
    cv::setIdentity(KF.measurementNoiseCov, cv::Scalar::all(10));
    cv::setIdentity(KF.errorCovPost, cv::Scalar::all(.1));

    // State Transition Matrix A
    // [1 0 0 dT 0 0]
    // [0 1 0 0 dT 0]
    // [0 0 1 0 0 dT]
    // [0 0 0 1 0 0]
    // [0 0 0 0 1 0]
    // [0 0 0 0 0 1] 
    KF.transitionMatrix = cv::Mat::zeros(cv::Size(STATE_SIZE, STATE_SIZE), CV_32FC1);
    KF.transitionMatrix.at<float>(0) = 1.0;
    KF.transitionMatrix.at<float>(3) = dT;
    KF.transitionMatrix.at<float>(7) = 1.0;
    KF.transitionMatrix.at<float>(10) = dT;
    KF.transitionMatrix.at<float>(14) = 1.0;
    KF.transitionMatrix.at<float>(17) = dT;
    KF.transitionMatrix.at<float>(21) = 1.0;
    KF.transitionMatrix.at<float>(28) = 1.0;
    KF.transitionMatrix.at<float>(35) = 1.0;

    // Control Input Matrix B
    // [dT^2/2 0 0]
    // [0 dT^2/2 0]
    // [0 0 dT^2/2]
    // [dT 0 0]
    // [0 dT 0]
    // [0 0 dT]
    KF.controlMatrix = cv::Mat::zeros(cv::Size(STATE_SIZE, MEAS_SIZE), CV_32FC1);
    KF.controlMatrix.at<float>(0) = pow(dT,2)/2;
    KF.controlMatrix.at<float>(4) = pow(dT,2)/2;
    KF.controlMatrix.at<float>(8) = pow(dT,2)/2;
    KF.controlMatrix.at<float>(9) = dT;
    KF.controlMatrix.at<float>(13) = dT;
    KF.controlMatrix.at<float>(17) = dT;

    // Measurement Mapping Transformation Matrix H
    // [1 0 0 0 0 0]
    // [0 1 0 0 0 0]
    // [0 0 1 0 0 0]
    KF.measurementMatrix = cv::Mat::zeros(cv::Size(MEAS_SIZE, STATE_SIZE), CV_32FC1);
    KF.measurementMatrix.at<float>(0) = 1.0;
    KF.measurementMatrix.at<float>(7) = 1.0;
    KF.measurementMatrix.at<float>(14) = 1.0;

    // Process Noise Covariance Matrix Q
    // [dT^4/4 0 0 dT^3/2 0 0]
    // [0 dT^4/4 0 0 dT^3/2 0]
    // [0 0 dT^4/4 0 0 dT^3/2]
    // [dT^3/2 0 0 dT^2 0 0]
    // [0 dT^3/2 0 0 dT^2 0]
    // [0 0 dT^3/2 0 0 dT^2]
    KF.processNoiseCov = cv::Mat::zeros(cv::Size(STATE_SIZE, STATE_SIZE), CV_32FC1);
    KF.processNoiseCov.at<float>(0) = SIGMA_A2*pow(dT,4)/4;
    KF.processNoiseCov.at<float>(3) = SIGMA_A2*pow(dT,3)/2;
    KF.processNoiseCov.at<float>(7) = SIGMA_A2*pow(dT,4)/4;
    KF.processNoiseCov.at<float>(10) = SIGMA_A2*pow(dT,3)/2;
    KF.processNoiseCov.at<float>(14) = SIGMA_A2*pow(dT,4)/4;
    KF.processNoiseCov.at<float>(17) = SIGMA_A2*pow(dT,3)/2;
    KF.processNoiseCov.at<float>(18) = SIGMA_A2*pow(dT,3)/2;
    KF.processNoiseCov.at<float>(21) = SIGMA_A2*pow(dT,2);
    KF.processNoiseCov.at<float>(25) = SIGMA_A2*pow(dT,3)/2;
    KF.processNoiseCov.at<float>(28) = SIGMA_A2*pow(dT,2);
    KF.processNoiseCov.at<float>(32) = SIGMA_A2*pow(dT,3)/2;
    KF.processNoiseCov.at<float>(35) = SIGMA_A2*pow(dT,2);

    // Measurement Noise Covariance Matrix R
    // [SIGMA_X 0 0]
    // [0 SIGMA_Y 0]
    // [0 0 SIGMA_Z] 
    KF.measurementNoiseCov = cv::Mat::zeros(cv::Size(MEAS_SIZE, MEAS_SIZE), CV_32FC1);
    KF.measurementNoiseCov.at<float>(0) = SIGMA_X;
    KF.measurementNoiseCov.at<float>(4) = SIGMA_Y;
    KF.measurementNoiseCov.at<float>(8) = SIGMA_Z;
    cv::setIdentity(KF.measurementNoiseCov, SIGMA_A2);

    // Initial Covariance Matrix P
    // >>> to initialize as Identity Matrix
    // KF.errorCovPost = cv::Mat::zeros(cv::Size(STATE_SIZE, STATE_SIZE), CV_32FC1);
    cv::setIdentity(KF.errorCovPost, 1.0);

    cv::Mat prediction = KF.predict();

    cv::Mat boat_meas = cv::Mat::zeros(cv::Size(MEAS_SIZE, 1), CV_32FC1);
    boat_meas.at<float>(0) = odometry.x;
    boat_meas.at<float>(1) = odometry.y;
    boat_meas.at<float>(2) = odometry.z;
    
    cv::Mat estimation = KF.correct(boat_meas);

```

```auto
OpenCV Error: Assertion failed (a_size.width == len) in gemm, file /build/opencv-L2vuMj/opencv-3.2.0+dfsg/modules/core/src/matmul.cpp, line 1537
terminate called after throwing an instance of 'cv::Exception'
  what(): /build/opencv-L2vuMj/opencv-3.2.0+dfsg/modules/core/src/matmul.cpp:1537: error: (-215) a_size.width == len in function gemm

Aborted

```

---

<div class="post-metadata">

**Author:** ![crackwitz](https://sea2.discourse-cdn.com/flex020/user_avatar/forum.opencv.org/crackwitz/32/14_2.png) [@crackwitz](https://forum.opencv.org/u/crackwitz)\
**Post date:** [December 16, 2021, 1:28pm UTC](https://forum.opencv.org/t/kalman-filter-with-opencv-c-assertion-failed-a-size-width-len-in-gemm/6767/2 "2021-12-16T13:28:14Z")

</div>

> [@diogofd8](#):
>
> `opencv-3.2.0`

please try again with a recent version. if this is a bug, in v3.2.0, it might be fixed. then any time spent on this would be wasted.

---

<div class="post-metadata">

**Author:** ![berak](https://avatars.discourse-cdn.com/v4/letter/b/85f322/32.png) [@berak](https://forum.opencv.org/u/berak)\
**Post date:** [December 16, 2021, 1:41pm UTC](https://forum.opencv.org/t/kalman-filter-with-opencv-c-assertion-failed-a-size-width-len-in-gemm/6767/3 "2021-12-16T13:41:58Z")

</div>

> [@diogofd8](#):
>
> I’m confident that I’m setting them right

however, one of them still has the wrong shape

debug it, get a backtrace (bt in gdb) , then you know, which of (sadly a lot)  
the matrix multiplications to check.

and maybe you can improve the code example above, to make it reproducable for others

---

<div class="post-metadata">

**Author:** ![diogofd8](https://avatars.discourse-cdn.com/v4/letter/d/bc8723/32.png) [@diogofd8](https://forum.opencv.org/u/diogofd8)\
**Post date:** [December 16, 2021, 8:09pm UTC](https://forum.opencv.org/t/kalman-filter-with-opencv-c-assertion-failed-a-size-width-len-in-gemm/6767/4 "2021-12-16T20:09:11Z")

</div>

I hate that I have to use ROS for this project and I’m very uncomfortable with it. I have no clue on how to use backtrace with it. I also don’t think I can update openCV. I know that the gazebo simulator I’m used needs ros melocic to run, on noetic it doesn’t behave as expected. I didn’t create this model, I just have to get the position from it and work on object tracking.

However, I tried printing the matrix sizes (with size()) and content.

Kalman Filter Init → dt = 0.322711 , measurements: [12.9959, -11.8709, -0.0850451]  
predict 1 \> [0, 0, 0] → this is the return of KF.predict()

```auto
KF.statePre: [1 x 6]
[0;
 0;
 0;
 0;
 0;
 0]

KF.statePost: [1 x 6]
[0;
 0;
 0;
 0;
 0;
 0]

KF.transitionMatrix: [6 x 6]
[1, 0, 0, 0.32271138, 0, 0;
 0, 1, 0, 0, 0.32271138, 0;
 0, 0, 1, 0, 0, 0.32271138;
 0, 0, 0, 1, 0, 0;
 0, 0, 0, 0, 1, 0;
 0, 0, 0, 0, 0, 1]

KF.controlMatrix: [3 x 6]
[0.052071314, 0, 0;
 0, 0.052071314, 0;
 0, 0, 0.052071314;
 0.32271138, 0, 0;
 0, 0.32271138, 0;
 0, 0, 0.32271138]

KF.measurementMatrix: [6 x 3]
[1, 0, 0, 0, 0, 0;
 0, 1, 0, 0, 0, 0;
 0, 0, 1, 0, 0, 0]

KF.processNoiseCov: [6 x 6]
[0.00027114217, 0, 0, 0.0016804005, 0, 0;
 0, 0.00027114217, 0, 0, 0.0016804005, 0;
 0, 0, 0.00027114217, 0, 0, 0.0016804005;
 0.0016804005, 0, 0, 0.010414262, 0, 0;
 0, 0.0016804005, 0, 0, 0.010414262, 0;
 0, 0, 0.0016804005, 0, 0, 0.010414262]

KF.measurementNoiseCov: [3 x 3]
[0.1, 0, 0;
 0, 0.1, 0;
 0, 0, 0.1]

KF.errorCovPost: [6 x 6]
[1.1044137, 0, 0, 0.32439178, 0, 0;
 0, 1.1044137, 0, 0, 0.32439178, 0;
 0, 0, 1.1044137, 0, 0, 0.32439178;
 0.32439178, 0, 0, 1.0104142, 0, 0;
 0, 0.32439178, 0, 0, 1.0104142, 0;
 0, 0, 0.32439178, 0, 0, 1.0104142]

KF.errorCovPre: [6 x 6]
[1.1044137, 0, 0, 0.32439178, 0, 0;
 0, 1.1044137, 0, 0, 0.32439178, 0;
 0, 0, 1.1044137, 0, 0, 0.32439178;
 0.32439178, 0, 0, 1.0104142, 0, 0;
 0, 0.32439178, 0, 0, 1.0104142, 0;
 0, 0, 0.32439178, 0, 0, 1.0104142]

KF.gain: [3 x 6]
[0, 0, 0;
 0, 0, 0;
 0, 0, 0;
 0, 0, 0;
 0, 0, 0;
 0, 0, 0]

```

```auto
OpenCV Error: Assertion failed (C.type() == type && (((flags&GEMM_3_T) == 0 && C.rows == d_size.height && C.cols == d_size.width) || ((flags&GEMM_3_T) != 0 && C.rows == d_size.width && C.cols == d_size.height))) in gemm, file /build/opencv-L2vuMj/opencv-3.2.0+dfsg/modules/core/src/matmul.cpp, line 1560
terminate called after throwing an instance of 'cv::Exception'
  what(): /build/opencv-L2vuMj/opencv-3.2.0+dfsg/modules/core/src/matmul.cpp:1560: error: (-215) C.type() == type && (((flags&GEMM_3_T) == 0 && C.rows == d_size.height && C.cols == d_size.width) || ((flags&GEMM_3_T) != 0 && C.rows == d_size.width && C.cols == d_size.height)) in function gemm

```